CN116226778B

CN116226778B - Retaining wall structure anomaly analysis method and system based on three-dimensional analysis platform

Info

Publication number: CN116226778B
Application number: CN202310516271.1A
Authority: CN
Inventors: 张海发; 郭林林; 肖尧轩; 杨猛; 钟志云; 陈翔; 农珊; 党宁; 许宏燕; 刘亚军; 谭森明; 赵博华
Original assignee: Zhujiang Water Resources Comprehensive Technology Center Of Zhujiang Water Resources Commission Of Ministry Of Water Resources
Current assignee: Zhujiang Water Resources Comprehensive Technology Center Of Zhujiang Water Resources Commission Of Ministry Of Water Resources
Priority date: 2023-05-09
Filing date: 2023-05-09
Publication date: 2023-07-07
Anticipated expiration: 2043-05-09
Also published as: CN116226778A

Abstract

The application relates to the technical field of three-dimensional analysis platforms and retaining wall structure abnormality analysis, and relates to a retaining wall structure abnormality analysis method and system based on a three-dimensional analysis platform. According to the method, based on the three-dimensional analysis platform, the characteristic domain of the retaining wall structure data corresponding to the abnormal characteristic embedded model meeting the model convergence condition is constructed, the characteristic domain of the obtained retaining wall structure data is migrated to the abnormal characteristic embedded model to be trained, and the converged first target abnormal characteristic embedded model is generated, so that independent sample data and training label data are not needed for learning, the model learning speed can be improved, the speed of classifying abnormal categories of the subsequent target retaining wall structure data is improved, the accuracy of classifying the retaining wall structure abnormality analysis is improved, and the efficiency of analyzing the retaining wall structure abnormality is improved.

Description

Retaining wall structure anomaly analysis method and system based on three-dimensional analysis platform

Technical Field

The application relates to the technical field of three-dimensional analysis platforms and retaining wall structure anomaly analysis, in particular to a retaining wall structure anomaly analysis method and system based on a three-dimensional analysis platform.

Background

The retaining wall is a structure for supporting roadbed filling soil or hillside soil and preventing deformation and instability of the filling soil or soil, various design configuration parameters of the retaining wall are loaded into the three-dimensional analysis platform at present, so that corresponding retaining wall structure data can be generated through the three-dimensional analysis platform, further, abnormal analysis of the retaining wall structure is conveniently carried out through computer equipment, and compared with on-site manual experience analysis, the analysis speed and the analysis quality are improved, and subsequent backtracking is convenient. In the related art, a neural network model is generally used for anomaly analysis, however, in the related art, when model training is performed, separate sample data and training label data are generally required for learning, and the model cannot be assisted to train by means of the existing model meeting the model convergence condition, so that the model learning speed is slower, and the speed of classifying the subsequent anomaly categories is further influenced.

Disclosure of Invention

In order to at least overcome the above-mentioned shortcomings in the prior art, an object of an embodiment of the present application is to provide a method and a system for analyzing an abnormality of a retaining wall structure based on a three-dimensional analysis platform.

In a first aspect, the present application provides a retaining wall structure anomaly analysis method based on a three-dimensional analysis platform, which is applied to a retaining wall structure anomaly analysis system based on a three-dimensional analysis platform, and the method includes:

Obtaining a target training data set, wherein the target training data set is obtained by sampling a to-be-learned retaining wall structure data set collected in a previous experiment, and each to-be-learned retaining wall structure data in the to-be-learned retaining wall structure data set is retaining wall structure data of marked abnormal label vectors generated based on a three-dimensional analysis platform;

respectively inputting each retaining wall structure data to be learned in the target training data set into an abnormal feature embedding model meeting model convergence conditions and an abnormal feature embedding model to be trained which is to be subjected to model convergence optimization to perform abnormal feature embedding, generating learned abnormal features corresponding to each retaining wall structure data to be learned and abnormal features to be learned corresponding to each retaining wall structure data to be learned, wherein the abnormal feature embedding model to be trained is generated by initializing and loading model weight information of the abnormal feature embedding model meeting model convergence conditions;

acquiring feature distances among learned abnormal features corresponding to the retaining wall structure data to be learned, generating first feature distance distribution, and acquiring feature distances among the abnormal features to be learned corresponding to the retaining wall structure data to be learned, generating second feature distance distribution;

Acquiring training effect parameter values between the second characteristic distance distribution and the first characteristic distance distribution, updating the to-be-trained abnormal feature embedded model waiting for model convergence optimization according to the training effect parameter values, returning to the step of acquiring a target training data set for iterative learning until the model convergence condition is met, outputting the converged to-be-trained abnormal feature embedded model as a first target abnormal feature embedded model, wherein the first target abnormal feature embedded model is used for extracting abnormal embedded features of target retaining wall structure data;

and carrying out abnormal category classification according to the abnormal embedded characteristics of the target retaining wall structure data.

In a possible implementation manner of the first aspect, the obtaining feature distances between learned abnormal features corresponding to the to-be-learned retaining wall structure data, generating a first feature distance distribution includes:

obtaining a first abnormal feature training cluster according to the learned abnormal features corresponding to the retaining wall structure data to be learned, and carrying out regularization conversion on the first abnormal feature training cluster to generate a regularized first abnormal feature training cluster;

Obtaining a scrambling feature training cluster corresponding to the regularized first abnormal feature training cluster, and generating a regularized second scrambling feature training cluster;

obtaining the first feature distance distribution according to the regularized second scrambling feature training cluster and the regularized first abnormal feature training cluster;

the obtaining the feature distance between the to-be-learned abnormal features corresponding to the to-be-learned retaining wall structure data, and generating a second feature distance distribution, includes:

obtaining a third abnormal feature training cluster according to the abnormal features to be learned corresponding to the retaining wall structure data to be learned, and carrying out regularization conversion on the third abnormal feature training cluster to generate a regularized third abnormal feature training cluster;

obtaining a scrambling feature training cluster corresponding to the regularized third abnormal feature training cluster, and generating a regularized fourth abnormal feature training cluster;

and obtaining the second characteristic distance distribution according to the regularized fourth abnormal characteristic training cluster and the regularized third abnormal characteristic training cluster.

In a possible implementation manner of the first aspect, the step of obtaining the training effect parameter value between the second feature distance distribution and the first feature distance distribution, and returning to the step of obtaining the target training data set for iterative learning after updating the to-be-trained abnormal feature embedding model waiting for model convergence optimization according to the training effect parameter value includes:

Obtaining standard deviation parameters of the second characteristic distance distribution and the first characteristic distance distribution, generating a first learning parameter value, and outputting the first learning parameter value as the training effect parameter value;

according to the training effect parameter value back propagation, updating the model weight information in the to-be-trained abnormal feature embedded model, and generating an updated abnormal feature embedded model;

outputting the updated abnormal characteristic embedded model as an abnormal characteristic embedded model to be trained, and returning to the step of acquiring the target training data set for iterative learning;

the obtaining the standard deviation parameter of the second characteristic distance distribution and the first characteristic distance distribution, and generating the training effect parameter value includes:

obtaining standard deviation parameters of the second characteristic distance distribution and the first characteristic distance distribution, and generating a first learning parameter value;

acquiring the learning data quantity corresponding to the target training data set, and acquiring the quotient of the first learning parameter value and the learning data quantity to acquire a second learning parameter value;

acquiring a preset influence coefficient, and carrying out weight fusion on the second learning parameter value according to the preset influence coefficient to generate a third learning parameter value;

And acquiring a fourth learning parameter value corresponding to the abnormal feature embedding model to be trained waiting for model convergence optimization, and acquiring a weighted fusion parameter value of the fourth learning parameter value and the third learning parameter value to generate the training effect parameter value.

In a possible implementation manner of the first aspect, the obtaining the training effect parameter value between the second feature distance distribution and the first feature distance distribution, updating the to-be-trained abnormal feature embedded model waiting for model convergence optimization according to the training effect parameter value, and returning to the step of obtaining the target training data set to perform iterative learning until the model convergence condition is met, and outputting the converged to-be-trained abnormal feature embedded model as the first target abnormal feature embedded model, where the method includes:

inputting the second characteristic distance distribution into a conversion network for parameter conversion to generate a target conversion characteristic distance distribution;

acquiring standard deviation parameters between the target conversion characteristic distance distribution and the first characteristic distance distribution, generating a target training effect parameter value, updating the conversion network and the abnormal characteristic embedding model to be trained according to the reverse propagation of the target training effect parameter value, and generating an updated conversion network and an updated abnormal characteristic embedding model;

Outputting the updated conversion network as a conversion network, outputting the updated abnormal feature embedded model as an abnormal feature embedded model to be trained, and returning to the step of acquiring the target training data set to perform iterative learning until the model convergence condition is met, and acquiring a second target abnormal feature embedded model through the converged abnormal feature embedded model to be trained and the converged conversion network.

inputting the second characteristic distance distribution and the first characteristic distance distribution into a target estimation network for category estimation, and generating characteristic distance estimation information;

and after updating the target estimation network and the abnormal feature embedding model to be trained according to the feature distance estimation information, returning to the step of acquiring the target training data set to perform iterative learning, and outputting the converged abnormal feature embedding model to be trained as a third target abnormal feature embedding model when the model convergence condition is met.

In a possible implementation manner of the first aspect, the target training dataset includes a plurality of multi-modal retaining wall structure data clusters including associated retaining wall structure data sets therein;

the method further comprises the steps of:

respectively inputting each multi-mode retaining wall structure data cluster into an abnormal feature embedding model meeting model convergence conditions and an abnormal feature embedding model to be trained waiting for model convergence optimization to perform abnormal feature embedding, and generating a first multi-mode abnormal feature corresponding to each multi-mode retaining wall structure data cluster and a second multi-mode abnormal feature corresponding to each multi-mode retaining wall structure data cluster;

and carrying out loss function value calculation according to the first multi-mode abnormal feature and the second multi-mode abnormal feature, generating first loss function value data, carrying out iterative learning according to the step of returning to the step of acquiring the target training data set after updating the to-be-trained abnormal feature embedded model according to the first loss function value data in a back propagation manner, and outputting the converged to-be-trained abnormal feature embedded model as a fourth target abnormal feature embedded model until the model convergence condition is met.

In a possible implementation manner of the first aspect, after the obtaining a training effect parameter value between the second feature distance distribution and the first feature distance distribution, updating the to-be-trained abnormal feature embedded model waiting for model convergence optimization according to the training effect parameter value, returning to the step of obtaining the target training data set to perform iterative learning until a model convergence condition is met, outputting the converged to-be-trained abnormal feature embedded model as a first target abnormal feature embedded model, and further including:

acquiring target retaining wall structure data, inputting the target retaining wall structure data into the first target abnormal feature embedding model to embed abnormal features, and generating abnormal embedded features to be analyzed;

acquiring prior abnormal category characteristics corresponding to prior abnormal category label data, and acquiring characteristic distances between the to-be-analyzed abnormal embedded characteristics and the prior abnormal category characteristics;

and determining abnormal category classification information corresponding to the target retaining wall structure data according to the characteristic distance.

In a possible implementation manner of the first aspect, the method further includes:

Acquiring retaining wall structure data to be excavated and excavated retaining wall structure data clusters;

inputting the retaining wall structure data to be excavated and the excavated retaining wall structure data cluster into an abnormal feature embedding model meeting model convergence conditions to perform abnormal feature embedding, generating an abnormal feature to be excavated corresponding to the retaining wall structure data to be excavated and an abnormal feature to be excavated corresponding to the retaining wall structure data cluster to be excavated, and obtaining feature distances of the abnormal feature to be excavated and the abnormal feature to be excavated, so as to generate a first feature distance distribution;

inputting the retaining wall structure data to be excavated and the excavated retaining wall structure data cluster into a target abnormal feature embedding model to perform abnormal feature embedding, generating an excavated abnormal embedding feature distribution corresponding to the retaining wall structure data to be excavated and the excavated abnormal embedding feature distribution corresponding to the retaining wall structure data cluster to be excavated, and obtaining feature distances between the target abnormal embedding feature to be excavated and the excavated abnormal embedding feature distribution to generate a second feature distance distribution, wherein the target abnormal feature embedding model is generated by performing migration parameter training based on an abnormal feature embedding model meeting model convergence conditions;

Performing correlation analysis according to the first characteristic distance distribution and the second characteristic distance distribution, and generating correlation parameter values corresponding to the retaining wall structure data to be excavated;

the obtaining the feature distance between the to-be-mined abnormal embedded feature and the mined abnormal embedded feature to generate a first feature distance distribution includes:

performing regular conversion on the to-be-mined abnormal embedded features to generate regular converted to-be-mined abnormal embedded features, and performing regular conversion on the mined abnormal embedded features to generate regular converted mined abnormal embedded features;

scrambling the regularly converted mined abnormal embedded features to generate a mined and scrambled feature training cluster, and obtaining the first feature distance distribution according to feature distances between the regularly converted mined abnormal embedded features and the mined and scrambled feature training cluster.

In a possible implementation manner of the first aspect, the obtaining the feature distance between the target to be mined abnormal embedded feature and the mined abnormal embedded feature distribution, and generating a second feature distance distribution include:

Performing regular conversion on the target abnormal embedded feature to be mined to generate regular conversion on the target abnormal embedded feature to be mined, and performing regular conversion on the mined abnormal embedded feature distribution to generate a mined abnormal embedded feature distribution after the regular conversion;

scrambling the regularly converted mined abnormal embedded feature distribution to generate a mined target scrambling feature training cluster, and obtaining the second feature distance distribution according to the regularly converted mined abnormal embedded feature distribution and the mined target scrambling feature training cluster;

performing correlation analysis according to the first characteristic distance distribution and the second characteristic distance distribution, generating a correlation parameter value corresponding to the retaining wall structure data to be excavated, and determining a commonality measurement correlation parameter value corresponding to the retaining wall structure data to be excavated according to the correlation parameter value corresponding to the retaining wall structure data to be excavated, including:

obtaining standard deviation parameters between the first characteristic distance distribution and the second characteristic distance distribution, and generating a target learning effect value;

determining the quantity of the retaining wall structure data to be excavated and the quantity of the retaining wall structure data corresponding to the excavated retaining wall structure data clusters, obtaining the quotient of the target learning effect value and the quantity of the retaining wall structure data, and determining the correlation parameter value corresponding to the retaining wall structure data to be excavated according to the quotient;

And if the correlation parameter value is larger than the threshold parameter value, obtaining the correlation estimation information corresponding to the retaining wall structure data to be excavated.

In a second aspect, the embodiment of the present application further provides a retaining wall structure anomaly analysis system based on a three-dimensional analysis platform, where the retaining wall structure anomaly analysis system based on a three-dimensional analysis platform includes a processor and a machine-readable storage medium, where machine-executable instructions are stored in the machine-readable storage medium, where the machine-executable instructions are loaded and executed by the processor to implement a retaining wall structure anomaly analysis method based on a three-dimensional analysis platform in any one of possible embodiments of the first aspect.

In any aspect of the above, the method includes respectively inputting each retaining wall structure data to be learned in a target training data set into an abnormal feature embedding model satisfying model convergence conditions and an abnormal feature embedding model to be trained for model convergence optimization, generating each abnormal feature to be learned corresponding to each retaining wall structure data to be learned and each abnormal feature to be learned corresponding to each retaining wall structure data to be learned, obtaining feature distances between the abnormal features to be learned corresponding to each retaining wall structure data to be learned, generating a first feature distance distribution, obtaining feature distances between the abnormal features to be learned corresponding to each retaining wall structure data to be learned, generating a second feature distance distribution, finally, based on training effect parameter values between the second feature distance distribution and the first feature distance distribution, optimizing the abnormal feature embedding model to be trained for model convergence optimization by the training effect parameter values, returning to the step of obtaining the target training data set for iterative learning until the model convergence conditions are satisfied, outputting the abnormal feature embedding model to be learned as a first target abnormal feature embedding model, building feature distances between the abnormal feature embedding models corresponding to be learned, generating a second feature distance distribution, and finally, and based on the training effect parameter values between the second feature distance distribution and the first feature distance distribution, improving the abnormal feature distance distribution, and the abnormal feature data can be trained by the training label model, and improving the abnormal feature data can be trained by the abnormal feature model after the abnormal feature model is required to be classified and the abnormal feature data is generated by the abnormal training model, and the abnormal feature model is required to be deformed and the abnormal feature-classified and is improved by the abnormal feature model, thereby improving the analysis efficiency of the retaining wall structure abnormality.

Drawings

For a clearer description of the technical solutions of the embodiments of the present application, reference will be made to the accompanying drawings, which are needed to be activated, for the sake of simplicity, and it should be understood that the following drawings only illustrate some embodiments of the present application, and therefore should not be considered as limiting the scope, and that it is possible for a person skilled in the art to extract other relevant drawings in combination with these drawings without the inventive effort.

Fig. 1 is a flow chart of a method for analyzing an abnormality of a retaining wall structure based on a three-dimensional analysis platform according to an embodiment of the present application;

fig. 2 is a schematic block diagram of a retaining wall structure anomaly analysis system based on a three-dimensional analysis platform for implementing the retaining wall structure anomaly analysis method based on a three-dimensional analysis platform according to an embodiment of the present application.

Detailed Description

The following description is presented to enable one of ordinary skill in the art to make and use the application and is provided in the context of a particular application and its requirements. It will be apparent to those having ordinary skill in the art that various changes can be made to the disclosed embodiments and that the general principles defined herein may be applied to other embodiments and applications without departing from the principles and scope of the present application. Thus, the present application is not limited to the embodiments described, but is to be accorded the widest scope consistent with the claims.

The following description is provided in connection with the accompanying drawings, and the specific operation method in the method embodiment may also be applied to the device embodiment or the system embodiment.

Fig. 1 is a schematic flow chart of a retaining wall structure abnormality analysis method based on a three-dimensional analysis platform according to an embodiment of the present application, and the retaining wall structure abnormality analysis method based on the three-dimensional analysis platform is described in detail below.

step101, obtaining a target training data set, wherein the target training data set is obtained by sampling a retaining wall structure data set to be learned collected in a previous experiment.

The target training dataset includes a plurality of soil-wall structure data to be learned at present, the soil-wall structure data to be learned at present representing soil-wall structure data to be taken at the time of the current training model. Each retaining wall structure data to be learned in the retaining wall structure data set is retaining wall structure data based on marked abnormal label vectors generated by the three-dimensional analysis platform. The prior collected retaining wall structure data set to be learned is a retaining wall structure data cluster of retaining wall structure data to be learned, which is adopted in the training process, and the target training data set is a part of the prior collected retaining wall structure data set to be learned. The retaining wall structure data to be learned in the retaining wall structure data set collected a priori is, for example, retaining wall structure data in which model learning is completed in an abnormal feature satisfying the model convergence condition. For example, a target training data set can be directly obtained by loading in a local or cloud end, wherein the target training data set is obtained by sampling in a retaining wall structure data set to be learned collected in a prior experiment.

step102, respectively inputting each retaining wall structure data to be learned in the target training data set into an abnormal feature embedding model meeting model convergence conditions and an abnormal feature embedding model to be trained waiting for model convergence optimization to perform abnormal feature embedding, generating learned abnormal features corresponding to each retaining wall structure data to be learned and abnormal features to be learned corresponding to each retaining wall structure data to be learned, wherein the abnormal feature embedding model to be trained is generated by initializing and loading model weight information of the abnormal feature embedding model meeting model convergence conditions.

The abnormal characteristic embedded model meeting the model convergence condition is an initialization model for mining the abnormal embedded characteristic, which is obtained after the model characteristic learning is carried out on the retaining wall structure data to be learned based on the history. It is currently necessary to learn the above initialized abnormal feature embedding model continuously to complete the update. The abnormal feature embedding model to be trained for model convergence optimization is generated by initializing the abnormal feature embedding model with model weight information to be updated or directly initializing and loading the model weight information of the converged abnormal feature embedding model. Then, the model weight information of the abnormal feature embedded model to be trained for model convergence optimization can be initialized based on the model weight information of the abnormal feature embedded model meeting the model convergence condition, and the model weight information of the abnormal feature embedded model meeting the model convergence condition is output as the initialized model weight information of the abnormal feature embedded model to be trained for model convergence optimization.

The learned abnormal characteristics are abnormal embedded characteristics of the retaining wall structure data to be learned obtained by mining the abnormal characteristic embedded model meeting model convergence conditions, and the abnormal characteristics to be learned are abnormal embedded characteristics of the retaining wall structure data to be learned obtained by carrying out abnormal characteristic embedding on the basis of the abnormal characteristic embedded model to be trained which is waiting for model convergence optimization. The abnormal embedded feature characterizes an abnormal feature vector representation of the retaining wall structure data.

The method includes the steps of inputting each retaining wall structure data to be learned in a target training data set into an abnormal feature embedding model meeting model convergence conditions to conduct abnormal feature embedding, generating learned abnormal features corresponding to each retaining wall structure data to be learned in the target training data set, inputting each retaining wall structure data to be learned in the target training data set into the abnormal feature embedding model to be trained to conduct model convergence optimization to conduct abnormal feature embedding, and generating abnormal features to be learned corresponding to each retaining wall structure data to be learned in the target training data set.

step103, obtaining feature distances among the learned abnormal features corresponding to the retaining wall structure data to be learned, generating a first feature distance distribution, and obtaining feature distances among the abnormal features to be learned corresponding to the retaining wall structure data to be learned, generating a second feature distance distribution.

For example, the first feature distance distribution includes each learned feature distance, the learned feature distance indicates a degree of association between learned abnormal features of two different retaining wall structure data to be learned, and feature distances between learned abnormal features corresponding to two pairs of retaining wall structure data to be learned in the target training data set are obtained. The second feature distance distribution comprises feature distances to be learned, and the feature distances to be learned indicate the association degree between abnormal features to be learned corresponding to two different retaining wall structure data to be learned. Traversing each retaining wall structure data to be learned in the target training data set, and acquiring the association degree between the retaining wall structure data to be learned currently and each retaining wall structure data to be learned in the target training data set. The method comprises the steps of obtaining learned feature distances based on learned abnormal feature acquisition, generating first feature distance distribution, and enabling an abnormal feature embedding model meeting model convergence conditions to represent an abnormal embedding feature domain of a target training data set obtained by abnormal feature embedding. Obtaining feature distance to be learned based on the feature distance to be learned, and generating second feature distance distribution, wherein the second feature distance distribution characterizes an abnormal embedding feature domain of a target training data set obtained by carrying out abnormal feature embedding on a to-be-trained abnormal feature embedding model waiting for model convergence optimization. In an alternative embodiment, the first characteristic distance distribution may be represented by a two-dimensional matrix and the second characteristic distance distribution may be the same.

step104, obtaining training effect parameter values between the second feature distance distribution and the first feature distance distribution, optimizing the to-be-trained abnormal feature embedded model waiting for model convergence optimization through the training effect parameter values, returning to the step of obtaining the target training data set for iterative learning until the converged to-be-trained abnormal feature embedded model is output as a first target abnormal feature embedded model when model convergence conditions are met, wherein the first target abnormal feature embedded model is used for extracting abnormal embedded features of target retaining wall structure data, and classifying abnormal categories through the abnormal embedded features of the target retaining wall structure data.

The training effect parameter value is indicative of a difference between the second characteristic distance distribution and the first characteristic distance distribution. For example, a difference between each feature distance to be learned in the second feature distance distribution and a corresponding learned feature distance in the first feature distance distribution may be obtained, and then a sum of all the differences may be obtained to obtain the training effect parameter value. And updating the model weight information in the abnormal feature embedding model to be trained waiting for model convergence optimization based on the training effect parameter value through back propagation, and generating an updated abnormal feature embedding model. And outputting the updated abnormal feature embedded model as an abnormal feature embedded model to be trained waiting for model convergence optimization, and performing iterative learning in the step of acquiring a subsequent target training data set until the model convergence condition is met, and outputting the converged abnormal feature embedded model to be trained as a first target abnormal feature embedded model. The first target abnormal feature embedding model is obtained after the to-be-trained abnormal feature embedding model waiting for model convergence optimization is learned, and is used for extracting the abnormal embedding features of the target retaining wall structure data, and carrying out abnormal category classification through the abnormal embedding features of the target retaining wall structure data.

Based on the above steps, each retaining wall structure data to be learned in the target training data set is respectively input into an abnormal feature embedding model meeting model convergence conditions and an abnormal feature embedding model to be trained which is to be subjected to model convergence optimization, abnormal feature embedding is carried out, learned abnormal features corresponding to each retaining wall structure data to be learned and abnormal features to be learned corresponding to each retaining wall structure data to be learned are generated, then feature distances between the learned abnormal features corresponding to each retaining wall structure data to be learned are acquired, a first feature distance distribution is generated, feature distances between the to-be-learned abnormal features corresponding to each retaining wall structure data to be learned are acquired, a second feature distance distribution is generated, finally training effect parameter values between the second feature distance distribution and the first feature distance distribution are obtained, after the model convergence optimization is carried out through the training effect parameter value optimization, iterative learning is carried out in the step of acquiring the target training data set until the model convergence conditions are met, the characteristic distance distribution is acquired, the retaining wall structure data to be learned is independent of the first target abnormal feature embedding model, the retaining wall data to be trained can be accurately transferred, the abnormal feature data can be accurately classified according to the training data is generated, the abnormal feature distance distribution is improved, the abnormal training data can be accurately transferred to the retaining wall structure is required to be embedded into the abnormal training model, and the abnormal feature data can be accurately classified, the abnormal model is required to be accurately changed, thereby improving the analysis efficiency of the retaining wall structure abnormality.

In an alternative embodiment, the step103 obtains feature distances between learned abnormal features corresponding to the structural data of each retaining wall to be learned, and generates a first feature distance distribution, which may be implemented by the following steps:

step1031, obtaining a first abnormal feature training cluster through the learned abnormal features corresponding to the retaining wall structure data to be learned, and performing regularization conversion on the first abnormal feature training cluster to generate a regularized first abnormal feature training cluster.

The first outlier feature training cluster may be a two-dimensional distribution matrix containing outlier embedded features of each learned outlier feature. The first abnormal feature training clusters are generated based on the learned abnormal features corresponding to the retaining wall structure data to be learned, and then the first abnormal feature training clusters are subjected to regularization conversion, for example, regularization conversion is completed based on a regularization conversion operator, for example, each component in the abnormal embedded features is divided by Euclidean norms of the abnormal embedded features, and the regularized first abnormal feature training clusters are generated.

step1032, obtaining a scrambling feature training cluster corresponding to the regularized first abnormal feature training cluster, generating a regularized second scrambling feature training cluster, and obtaining a first feature distance distribution through the regularized second scrambling feature training cluster and the regularized first abnormal feature training cluster.

In an alternative embodiment, the obtaining the first feature distance distribution corresponding to the regularized second scrambled feature training cluster and the regularized first abnormal feature training cluster may be obtaining the euclidean distance, the ming distance, the cosine distance, and the like corresponding to the regularized second scrambled feature training cluster and the regularized first abnormal feature training cluster, and generating the first feature distance distribution.

In an alternative embodiment, the step103 obtains feature distances between the abnormal features to be learned corresponding to the structural data of each retaining wall to be learned, and generates the second feature distance distribution, which may be implemented by the following steps:

step1033, obtaining a third abnormal feature training cluster through the abnormal features to be learned corresponding to the retaining wall structure data to be learned, and carrying out regularization conversion on the third abnormal feature training cluster to generate a regularized third abnormal feature training cluster.

step1034, obtaining a scrambling feature training cluster corresponding to the regularized third abnormal feature training cluster, generating a regularized fourth abnormal feature training cluster, and obtaining second feature distance distribution through the regularized fourth abnormal feature training cluster and the regularized third abnormal feature training cluster.

The procedure of step1033 and step1034 may refer to the principles of step1031 and step 1032.

In an alternative embodiment, the step104 obtains the training effect parameter value between the second feature distance distribution and the first feature distance distribution, and returns the step of obtaining the target training data set to perform iterative learning after the abnormal feature to be trained waiting for model convergence optimization is embedded into the model through the optimization of the training effect parameter value, which can be implemented by the following steps:

step1041, obtaining standard deviation parameters of the second feature distance distribution and the first feature distance distribution, generating a first learning parameter value, and outputting the first learning parameter value as a training effect parameter value.

step1042, through the back propagation of training effect parameter values, optimizes the model weight information in the to-be-trained abnormal feature embedded model, and generates an updated abnormal feature embedded model.

According to the embodiment of the application, the gradient is acquired through the training effect parameter value, then the gradient is reversely transmitted to the abnormal feature embedding model to be trained, optimization of model weight information in the abnormal feature embedding model to be trained is completed, and the updated abnormal feature embedding model is generated.

step1043, outputting the updated abnormal feature embedded model as the abnormal feature embedded model to be trained, and returning to the step of obtaining the target training data set for iterative learning.

In an alternative embodiment, the step1041 obtains the standard deviation parameter of the second feature distance distribution and the first feature distance distribution, and generates the training effect parameter value, which may be implemented by the following steps:

step10411, obtaining a standard deviation parameter of the second feature distance distribution and the first feature distance distribution, and generating a first learning parameter value.

For example, euclidean distances of feature distances to be learned in the second feature distance distribution and learned feature distances in the first feature distance distribution are obtained to obtain the first learning parameter value.

step10412, obtaining the learning data quantity corresponding to the target training data set, and obtaining the quotient of the first learning parameter value and the learning data quantity to obtain the second learning parameter value.

step10412, obtaining a preset influence coefficient, and performing weight fusion on the second learning parameter value through the preset influence coefficient to generate a third learning parameter value.

step10413, obtaining a fourth learning parameter value corresponding to the to-be-trained abnormal feature embedded model waiting for model convergence optimization, and obtaining a weighted fusion parameter value of the fourth learning parameter value and the third learning parameter value to generate a training effect parameter value.

And the fourth learning parameter value is a loss function value when the abnormal feature embedding model to be trained waiting for model convergence optimization performs abnormal category classification tasks after abnormal feature embedding.

In an alternative embodiment, in step104, training effect parameter values between the second feature distance distribution and the first feature distance distribution are obtained, after the to-be-trained abnormal feature embedded model waiting for model convergence optimization is optimized through the training effect parameter values, the step of returning to the step of obtaining the target training data set is performed for iterative learning, until the model convergence condition is met, and the converged to-be-trained abnormal feature embedded model is output as the first target abnormal feature embedded model, including:

(1) And inputting the second characteristic distance distribution into a conversion network to perform parameter conversion, and generating a target conversion characteristic distance distribution.

The conversion network is used for carrying out model weight information initialization, and is used for executing conversion on the second characteristic distance distribution so as to reduce characteristic gaps, the conversion network is a neural network, and the target conversion characteristic distance distribution is a sequence obtained after conversion. And inputting each feature distance to be learned in the second feature distance distribution into a conversion network to perform parameter conversion, and generating a target conversion feature distance distribution output by the conversion network.

(2) Standard deviation parameters between the target conversion characteristic distance distribution and the first characteristic distance distribution are obtained, a target training effect parameter value is generated, the conversion network and the abnormal characteristic embedding model to be trained are optimized through back propagation of the target training effect parameter value, and an updated conversion network and an updated abnormal characteristic embedding model are generated.

(3) Outputting the updated conversion network as a conversion network, outputting the updated abnormal feature embedded model as an abnormal feature embedded model to be trained, and returning to the step of acquiring the target training data set to perform iterative learning until the model convergence condition is met, and acquiring a second target abnormal feature embedded model through the converged abnormal feature embedded model to be trained and the converged conversion network.

For example, according to the target conversion feature distance distribution and the learned feature distance, obtaining the average distribution error to obtain the target training effect parameter value, then, based on the target training effect parameter value, performing back propagation to further optimize the model weight information of the conversion network and the model weight information in the abnormal feature embedding model to be trained, generating an updated conversion network and an updated abnormal feature embedding model, then, outputting the updated conversion network as the conversion network, outputting the updated abnormal feature embedding model as the abnormal feature embedding model to be trained, and returning to the step of obtaining the target training data set to perform iterative learning until the model convergence condition is met, and obtaining a second target abnormal feature embedding model through the converged abnormal feature embedding model to be trained and the converged conversion network. In other words, the second target abnormal feature embedding model includes a converged to-be-trained abnormal feature embedding model and a converged conversion network. According to the method and the device for embedding the abnormal characteristics to be trained, the conversion network is arranged on the abnormal characteristics to be trained embedded model, the abnormal characteristics to be trained embedded model and the conversion network are trained together, so that the second target abnormal characteristics embedded model is obtained, and the accuracy of abnormal characteristics embedding can be improved based on the second target abnormal characteristics embedded model.

In an alternative embodiment, the step104 obtains the training effect parameter value between the second feature distance distribution and the first feature distance distribution, and after the training effect parameter value is optimized and waiting for the to-be-trained abnormal feature embedded model subjected to model convergence optimization, returns to the step of obtaining the target training data set to perform iterative learning, until the model convergence condition is met, outputs the converged to-be-trained abnormal feature embedded model as the first target abnormal feature embedded model, and may be implemented by the following steps: and inputting the second characteristic distance distribution and the first characteristic distance distribution into a target estimation network for recognition, generating characteristic distance estimation information, optimizing the target estimation network and the abnormal characteristic embedding model to be trained through the characteristic distance estimation information, returning to the step of acquiring the target training data set for iterative learning until the convergence condition of the model is met, and outputting the converged abnormal characteristic embedding model to be trained as a third target abnormal characteristic embedding model.

In an alternative embodiment, the target training dataset may comprise a plurality of multi-modal retaining wall structure data clusters including associated retaining wall structure data sets therein; based on this, the above method further comprises:

step100, respectively inputting each multi-mode retaining wall structure data cluster into an abnormal feature embedding model meeting model convergence conditions and an abnormal feature embedding model waiting to be trained for model convergence optimization to perform abnormal feature embedding, and generating a first multi-mode abnormal feature corresponding to each multi-mode retaining wall structure data cluster and a second multi-mode abnormal feature corresponding to each multi-mode retaining wall structure data cluster.

The target training data set includes each multi-modal retaining wall structure data cluster including the associated retaining wall structure data set, in other words, two retaining wall structure data to be learned in the multi-modal retaining wall structure data cluster are retaining wall structure data of the same category, the other retaining wall structure data to be learned and the two retaining wall structure data of the same category are retaining wall structure data of different categories, the retaining wall structure data of the same category is a positive retaining wall structure data set, and one retaining wall structure data and the retaining wall structure data of different categories in the retaining wall structure data of the same category constitute a negative retaining wall structure data set. The multi-modal abnormal characteristics are obtained by embedding abnormal characteristics of each retaining wall structure data to be learned in the multi-modal retaining wall structure data cluster and then obtaining multi-modal abnormal characteristics obtained by constructing each characteristic. The abnormal embedded features of the data of the retaining wall to be learned are combined to obtain multi-mode abnormal features, and the first multi-mode abnormal features are obtained by embedding the abnormal features of the data clusters of the multi-mode retaining wall by an abnormal feature embedded model meeting model convergence conditions. The second multi-mode abnormal feature is a feature obtained by carrying out abnormal feature embedding on the multi-mode retaining wall structure data cluster by an abnormal feature embedding model to be trained waiting for model convergence optimization. For example, each multi-modal retaining wall structure data cluster is input into an abnormal feature embedding model meeting model convergence conditions and an abnormal feature embedding model to be trained waiting for model convergence optimization to perform abnormal feature embedding, a first multi-modal abnormal feature corresponding to each multi-modal retaining wall structure data cluster is generated, each multi-modal retaining wall structure data cluster is input into the abnormal feature embedding model to be trained waiting for model convergence optimization to perform abnormal feature embedding, and a second multi-modal abnormal feature corresponding to each multi-modal retaining wall structure data cluster is generated.

step200, performing loss function value calculation through the first multi-mode abnormal feature and the second multi-mode abnormal feature, generating first loss function value data, performing back propagation through the first loss function value data to optimize the abnormal feature embedded model to be trained, and returning to the step of obtaining the target training data set to perform iterative learning until the model convergence condition is met, and outputting the converged abnormal feature embedded model to be trained as a fourth target abnormal feature embedded model.

The first loss function value data indicates errors corresponding to the first multi-modal anomaly characteristic and the second multi-modal anomaly characteristic. For example, a loss function value calculation is performed based on the first multi-modal abnormal characteristics to generate a learned multi-modal set training effect parameter value, and meanwhile, a loss function value calculation is performed based on the multi-modal retaining wall structure data cluster to be learned to generate a multi-modal set training effect parameter value to be learned, and then a difference value between the learned multi-modal set training effect parameter value and the multi-modal set training effect parameter value to be learned is obtained to obtain first loss function value data. And then, carrying out back propagation through the first loss function value data so as to optimize the abnormal feature embedded model to be trained, and returning to the step of acquiring the target training data set to carry out iterative learning until the convergence condition of the model is met, and outputting the converged abnormal feature embedded model to be trained as a fourth target abnormal feature embedded model. The method comprises the steps of respectively inputting each multi-mode retaining wall structure data cluster into an abnormal feature embedding model meeting model convergence conditions and an abnormal feature embedding model to be trained waiting for model convergence optimization to conduct abnormal feature embedding, generating a first multi-mode abnormal feature corresponding to each multi-mode retaining wall structure data cluster and a second multi-mode abnormal feature corresponding to each multi-mode retaining wall structure data cluster, conducting loss function value calculation through the first multi-mode abnormal feature and the second multi-mode abnormal feature, generating first loss function value data, conducting back propagation through the first loss function value data to optimize the abnormal feature embedding model to be trained, returning to the step of acquiring a target training data set to conduct iterative learning until the converged abnormal feature embedding model to be trained is output as a fourth target abnormal feature embedding model when the model convergence conditions are met, and improving the performance of the obtained abnormal feature embedding model.

In an alternative embodiment, after obtaining the training effect parameter value between the second feature distance distribution and the first feature distance distribution, and optimizing the to-be-trained abnormal feature embedded model waiting for model convergence optimization through the training effect parameter value, returning to the step of obtaining the target training data set to perform iterative learning until the model convergence condition is met, outputting the converged to-be-trained abnormal feature embedded model as the first target abnormal feature embedded model, and further including:

step110, obtaining target retaining wall structure data, inputting the target retaining wall structure data into a first target abnormal feature embedding model for abnormal feature embedding, and generating abnormal embedded features to be analyzed.

The method comprises the steps that target retaining wall structure data are retaining wall structure data to be subjected to abnormal category analysis, abnormal embedding characteristics to be analyzed are abnormal embedding characteristics corresponding to the target retaining wall structure data, the target retaining wall structure data are input into a first target abnormal characteristic embedding model to be subjected to abnormal characteristic embedding, and abnormal embedding characteristics to be analyzed are generated.

step120, acquiring the prior abnormal category characteristics corresponding to the prior abnormal category label data, and acquiring the characteristic distance between the to-be-analyzed abnormal embedded characteristics and the prior abnormal category characteristics.

step130, determining abnormal category classification information corresponding to the target retaining wall structure data through the characteristic distance.

The prior abnormal category label data is used for representing prior retaining wall structure data and corresponding prior abnormal category characteristics. The prior retaining wall structure data are retaining wall structure data containing abnormal categories, and the prior abnormal category features are abnormal embedded features corresponding to the prior retaining wall structure data. For example, the prior abnormal category label data is collected in advance, and the prior abnormal category label data stores prior abnormal category characteristics corresponding to the prior retaining wall structure data. The prior abnormal category characteristics in the prior abnormal category label data are obtained, the feature distance between the to-be-analyzed abnormal embedded features and the prior abnormal category characteristics is obtained, and if the feature distance is larger than the preset feature distance, the corresponding abnormal category is contained in the representative target retaining wall structure data. At this time, the abnormality category classification information corresponding to the target retaining wall structure data is the target abnormality category identified. If the feature distance between the feature to be analyzed and each priori abnormal category feature in the priori abnormal category label data is smaller than the preset feature distance, the feature distance represents that no abnormal category information exists in the target retaining wall structure data, and the abnormal category classification information corresponding to the target retaining wall structure data is that no target abnormal category is identified. And carrying out abnormal feature embedding on the target retaining wall structure data through the first target abnormal feature embedding model to generate an abnormal embedding feature to be analyzed, then carrying out feature distance acquisition on the abnormal embedding feature to be analyzed and each priori retaining wall structure data in the priori abnormal class label data, and determining abnormal class classification information corresponding to the target retaining wall structure data through the feature distance, so that the speed of generating the abnormal class classification information is improved.

In an alternative embodiment, the above method may further comprise the steps of:

step210, obtaining retaining wall structure data to be excavated and excavated retaining wall structure data clusters.

In this embodiment, the excavated retaining wall structure data cluster includes a plurality of excavated retaining wall structure data for performing correlation analysis based on the abnormal feature embedding model satisfying the model convergence condition, and the retaining wall structure data to be excavated is retaining wall structure data to be subjected to correlation analysis, for example, retaining wall structure data in the excavated retaining wall structure data cluster. And performing iterative learning optimization on the abnormal characteristic embedded model meeting the model convergence condition, and then performing correlation analysis on the excavated retaining wall structure data again.

step220, inputting the retaining wall structure data to be excavated and the excavated retaining wall structure data cluster into an abnormal feature embedding model meeting the model convergence condition to perform abnormal feature embedding, generating an abnormal embedding feature to be excavated corresponding to the retaining wall structure data to be excavated and an abnormal embedding feature to be excavated corresponding to the retaining wall structure data cluster to be excavated, acquiring feature distances of the abnormal embedding feature to be excavated and the abnormal embedding feature to be excavated, and generating a first feature distance distribution.

The abnormal feature embedding model meeting the model convergence condition is an abnormal feature embedding model generated based on training of past retaining wall structure data to be learned, the first feature distance distribution comprises a plurality of first feature distances, the first feature distances indicate feature distances between the to-be-excavated abnormal embedding features and the excavated abnormal embedding features in the excavated abnormal embedding features, the to-be-excavated abnormal embedding features are abnormal embedding features corresponding to the retaining wall structure data to be excavated, and the excavated abnormal embedding features comprise a plurality of abnormal embedding features corresponding to the retaining wall structure data to be excavated. For example, an abnormal feature embedding model meeting the model convergence condition is called, the retaining wall structure data to be excavated and the excavated retaining wall structure data clusters are respectively input into the abnormal feature embedding model meeting the model convergence condition to conduct abnormal feature embedding, the to-be-excavated abnormal embedding feature corresponding to the retaining wall structure data to be excavated and the excavated abnormal embedding feature corresponding to the excavated retaining wall structure data clusters are generated, then feature distances of the to-be-excavated abnormal embedding feature and each excavated feature in the excavated abnormal embedding feature are obtained, and first feature distance distribution is generated.

step230, inputting the retaining wall structure data to be excavated and the data clusters of the retaining wall structure data to be excavated into a target abnormal feature embedding model to perform abnormal feature embedding, generating the target abnormal embedded feature to be excavated corresponding to the retaining wall structure data to be excavated and the abnormal embedded feature distribution to be excavated corresponding to the data clusters of the retaining wall structure data to be excavated, acquiring the feature distance between the target abnormal embedded feature to be excavated and the abnormal embedded feature distribution to be excavated, generating a second feature distance distribution, and performing migration parameter training on the basis of the abnormal feature embedding model meeting model convergence conditions to generate the target abnormal feature embedding model.

The target abnormal feature embedding model is generated by carrying out migration parameter training based on an abnormal feature embedding model meeting model convergence conditions, the target abnormal embedding feature to be mined is an abnormal embedding feature corresponding to the retaining wall structure data to be mined, which is obtained through extraction of the target abnormal feature embedding model, the mined abnormal embedding feature distribution comprises abnormal embedding features obtained through mining of the target abnormal feature embedding model of each piece of the retaining wall structure data to be mined, the second feature distance distribution comprises each second feature distance, and the second feature distance indicates feature distances between the target abnormal embedding feature to be mined and the target abnormal embedding feature to be mined in the mined abnormal embedding feature distribution.

For example, a target abnormal feature embedding model is called, the retaining wall structure data to be excavated and the excavated retaining wall structure data clusters are respectively input into the target abnormal feature embedding model to be embedded with abnormal features, the target abnormal embedded features to be excavated corresponding to the retaining wall structure data to be excavated and the excavated abnormal embedded feature distribution corresponding to the excavated retaining wall structure data clusters are generated, then feature distances between the target abnormal embedded features to be excavated and the excavated target features in the excavated abnormal embedded feature distribution are obtained, and a second feature distance distribution is generated.

step240, performing correlation analysis through the first characteristic distance distribution and the second characteristic distance distribution, and generating correlation parameter values corresponding to the retaining wall structure data to be excavated.

Therefore, through obtaining the retaining wall structure data to be excavated and the retaining wall structure data clusters to be excavated, inputting the retaining wall structure data to be excavated and the retaining wall structure data clusters to be excavated into the abnormal feature embedding model meeting the model convergence condition and the target abnormal feature embedding model for correlation analysis, because the target abnormal feature embedding model is generated by adopting the transfer learning based on the abnormal feature embedding model meeting the model convergence condition, the excavation efficiency of the retaining wall structure data to be excavated can be improved, meanwhile, the retaining wall structure data to be excavated and the retaining wall structure data clusters to be excavated are subjected to abnormal feature embedding based on the target abnormal feature embedding model and the abnormal feature embedding model meeting the model convergence condition, the first feature distance distribution and the second feature distance distribution are determined, and then correlation parameter values corresponding to the retaining wall structure data to be excavated are determined according to the correlation analysis of the first feature distance distribution and the second feature distance distribution, so that the reliability of the correlation parameter values is improved.

In an alternative embodiment, the feature distance between the feature to be mined and the feature of the mined feature to be embedded is obtained, and the first feature distance distribution is generated, which may be implemented by the following steps: the method comprises the steps of performing regular conversion on an abnormal embedded feature to be mined, generating a regular converted abnormal embedded feature to be mined, performing regular conversion on the abnormal embedded feature to be mined, generating a regular converted abnormal embedded feature to be mined, scrambling the regular converted abnormal embedded feature to be mined, generating a training cluster of the scrambling feature to be mined, and obtaining a first feature distance distribution through feature distances between the regular converted abnormal embedded feature to be mined and the training cluster of the scrambling feature to be mined.

The regularly converted abnormal embedded feature to be mined is the regularly converted abnormal embedded feature to be mined, and the training cluster of the mining and scrambling features is a training cluster of scrambling features generated by scrambling the regularly converted abnormal embedded feature to be mined.

In an alternative embodiment, the feature distance between the abnormal embedded feature of the object to be mined and the mined abnormal embedded feature distribution is obtained, and the second feature distance distribution is generated, which can be realized through the following steps: regularly converting the abnormal embedded features of the target to be mined to generate regularly converted abnormal embedded features of the target to be mined, regularly converting the distribution of the abnormal embedded features to generate regularly converted distribution of the abnormal embedded features; scrambling the regularly converted mined abnormal embedded feature distribution, generating a mined target scrambling feature training cluster, and obtaining a second feature distance distribution through the regularly converted mined abnormal embedded feature distribution and the mined target scrambling feature training cluster.

In an alternative embodiment, the correlation analysis is performed through the first characteristic distance distribution and the second characteristic distance distribution, the correlation parameter value corresponding to the retaining wall structure data to be excavated is generated, and the commonality measurement correlation parameter value corresponding to the retaining wall structure data to be excavated is determined through the correlation parameter value corresponding to the retaining wall structure data to be excavated, which can be achieved through the following steps:

(A) And acquiring standard deviation parameters between the first characteristic distance distribution and the second characteristic distance distribution, and generating a target learning effect value.

(B) And determining the quantity of the retaining wall structure data to be excavated and the quantity of the retaining wall structure data corresponding to the excavated retaining wall structure data clusters, obtaining the quotient of the target learning effect value and the quantity of the retaining wall structure data, and determining the correlation parameter value corresponding to the retaining wall structure data to be excavated through the quotient.

(C) And if the correlation parameter value is larger than the threshold parameter value, obtaining the correlation estimation information corresponding to the retaining wall structure data to be excavated.

The association estimation information represents that an abnormal embedded feature of the retaining wall structure data to be excavated may replace the same retaining wall structure data in the excavated retaining wall structure data cluster.

Fig. 2 illustrates a hardware structural intent of the three-dimensional analysis platform-based retaining wall structure abnormality analysis system 100 for implementing the three-dimensional analysis platform-based retaining wall structure abnormality analysis method provided in the embodiment of the present application, and as illustrated in fig. 2, the three-dimensional analysis platform-based retaining wall structure abnormality analysis system 100 may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

In an exemplary design concept, the retaining wall structure abnormality analysis system 100 based on the three-dimensional analysis platform may be a single retaining wall structure abnormality analysis system based on the three-dimensional analysis platform or a group of retaining wall structure abnormality analysis systems based on the three-dimensional analysis platform. The set of three-dimensional analysis platform-based retaining wall structure anomaly analysis systems may be centralized or distributed (e.g., the three-dimensional analysis platform-based retaining wall structure anomaly analysis system 100 may be a distributed system). In an exemplary design concept, the retaining wall structure anomaly analysis system 100 based on a three-dimensional analysis platform may be local or remote. For example, the retaining wall structure anomaly analysis system 100 based on a three-dimensional analysis platform may access information and/or data stored in the machine-readable storage medium 120 via a network. As another example, the retaining wall structure anomaly analysis system 100 based on a three-dimensional analysis platform may be directly connected to the machine-readable storage medium 120 to access stored information and/or data. In an exemplary design concept, the retaining wall structure anomaly analysis system 100 based on a three-dimensional analysis platform may be implemented on a cloud platform. For example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, or the like, or any combination thereof.

The machine-readable storage medium 120 may store data and/or instructions. In an exemplary design, machine-readable storage medium 120 may store data obtained from an external terminal. In an exemplary design concept, the machine-readable storage medium 120 may store data and/or instructions for use by the three-dimensional analysis platform-based retaining wall structure anomaly analysis system 100 to perform or use to complete the exemplary methods described herein. In an exemplary design, machine-readable storage medium 120 may include mass storage, removable storage, volatile read-write memory, read-only memory (ROM), and the like, or any combination thereof. Exemplary mass storage devices may include magnetic disks, optical disks, solid state disks, and the like. Exemplary removable memory may include flash drives, floppy disks, optical disks, memory cards, compact disks, tape, and the like. Exemplary volatile read-write memory can include Random Access Memory (RAM). Exemplary RAM may include active random access memory (DRAM), double data rate synchronous active random access memory (DDR SDRAM), passive random access memory (SRAM), thyristor random access memory (T-RAM), zero capacitance random access memory (Z-RAM), and the like. Exemplary read-only memory may include mask read-only memory (MROM), programmable read-only memory (PROM), erasable programmable read-only memory (PEROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM), digital versatile disk read-only memory, and the like. In an exemplary design, machine-readable storage medium 120 may be implemented on a cloud platform. For example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any combination thereof.

In a specific implementation process, at least one processor 110 executes computer executable instructions stored by the machine readable storage medium 120, so that the processor 110 may execute the retaining wall structure anomaly analysis method based on the three-dimensional analysis platform according to the above method embodiment, the processor 110, the machine readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 may be used to control the transceiving actions of the communication unit 140.

The specific implementation process of the processor 110 may refer to the above embodiments of the method executed by the retaining wall structure anomaly analysis system 100 based on the three-dimensional analysis platform, and the implementation principle and technical effects are similar, which is not described herein again.

In addition, the embodiment of the application also provides a readable storage medium, wherein computer executable instructions are preset in the readable storage medium, and when a processor executes the computer executable instructions, the retaining wall structure anomaly analysis method based on the three-dimensional analysis platform is realized.

It is to be understood that the above description is intended to be illustrative only and is not intended to limit the scope of the present application. Many modifications and variations will be apparent to those of ordinary skill in the art in light of the description herein. However, such modifications and variations do not depart from the scope of the present application.

While the basic concepts have been described above, it will be apparent to those of ordinary skill in the art after reading this application that the above disclosure is by way of example only and is not limiting of the present application. Although not explicitly described herein, various modifications, improvements, and adaptations of the present application may occur to one skilled in the art. Such modifications, improvements, and modifications are intended to be suggested within this application, and are therefore within the spirit and scope of the exemplary embodiments of this application.

Meanwhile, the present application uses specific words to describe embodiments of the present application. For example, "one embodiment," "an embodiment," and/or "some embodiments" means a particular feature, structure, or characteristic associated with at least one embodiment of the present application. Thus, it should be emphasized and should be appreciated that two or more references to "an embodiment" or "one embodiment" or "an alternative embodiment" in various positions in this specification are not necessarily referring to the same embodiment. Furthermore, certain features, structures, or characteristics of one or more embodiments of the present application may be combined as suitable.

Furthermore, those of ordinary skill in the art will appreciate that aspects of the invention are capable of being illustrated and described in connection with a variety of patentable categories or circumstances, including any novel and useful process, machine, product, or combination of materials, or any novel and useful modifications thereof. Accordingly, aspects of the present application may be performed entirely by hardware, entirely by software (including firmware, resident software, micro-code, etc.) or by a combination of hardware and software. The above hardware or software may be referred to as a "unit," module, "or" system. Furthermore, aspects of the present disclosure may take the form of a computer program product embodied in one or more computer-readable media, wherein the computer-readable program code is embodied therein.

The computer readable signal medium may comprise a propagated data signal with computer program code embodied therein, for example, on a baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including electro-magnetic, optical, etc., or any suitable combination thereof. A computer readable signal medium may be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code located on a computer readable signal medium may be propagated in connection with any suitable medium including radio, cable, fiber optic cable, RF, or the like, or any combination thereof.

The computer program code necessary for operation of portions of the present application may be written in any one or more programming languages, including a host-oriented programming language such as Java, scala, smalltalk, eiffel, JADE, emerald, C ++, c#, vb net, python, and the like, a conventional programming language such as C language, visual Basic, fortran 2003, perl, COBOL 2002, PHP, ABAP, an active programming language such as Python, ruby, and Groovy, or other programming languages, and the like. The program code may execute entirely on the computer or as a stand-alone software package, partly on the computer and partly on a remote computer or entirely on the remote computer or system. In the latter scenario, the remote computer may be connected to the computer in any network form, such as a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, in connection with the Internet), or the connection may be made to a cloud computing environment, or as a service, such as software as a service (SaaS).

Furthermore, the order in which the elements and sequences are presented, the use of numerical letters, or other designations are used in the application and are not intended to limit the order in which the processes and methods of the application are performed unless explicitly recited in the claims. While certain presently useful inventive embodiments have been discussed in connection with various examples thereof, it is to be understood that such detail is solely for that purpose and that the appended claims are not limited to the disclosed embodiments, but, on the contrary, are intended to cover all modifications and equivalent arrangements that are within the spirit and scope of the embodiments of the subject application. For example, while the system components described above may be implemented in connection with hardware devices, it may also be implemented in connection with software only solutions, such as installing the described system on an existing system or mobile device.

Likewise, it should be noted that in order to simplify the presentation disclosed herein and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof. Likewise, it should be noted that in order to simplify the presentation disclosed herein and thereby aid in understanding one or more inventive embodiments, various features are sometimes grouped together in a single embodiment, figure, or description thereof.

Claims

1. A retaining wall structure anomaly analysis method based on a three-dimensional analysis platform, characterized in that the method is realized by a retaining wall structure anomaly analysis system based on the three-dimensional analysis platform, and comprises the following steps:

2. The method for analyzing the abnormality of the retaining wall structure based on the three-dimensional analysis platform according to claim 1, wherein the obtaining the feature distance between the learned abnormality features corresponding to the respective retaining wall structure data to be learned, generating the first feature distance distribution, includes:

3. The method for analyzing the abnormality of the retaining wall structure based on the three-dimensional analysis platform according to claim 1, wherein the step of obtaining the training effect parameter value between the second characteristic distance distribution and the first characteristic distance distribution, updating the model to be trained for model convergence optimization according to the training effect parameter value, and returning to the step of obtaining the target training data set for iterative learning includes:

4. The method for analyzing the structural anomaly of a retaining wall based on a three-dimensional analysis platform according to claim 1, wherein the steps of obtaining the training effect parameter value between the second characteristic distance distribution and the first characteristic distance distribution, updating the to-be-trained anomaly characteristic embedded model waiting for model convergence optimization according to the training effect parameter value, and returning the obtained target training data set to perform iterative learning until the model convergence condition is satisfied, and outputting the converged to-be-trained anomaly characteristic embedded model as a first target anomaly characteristic embedded model, comprise:

5. The method for analyzing the structural anomaly of a retaining wall based on a three-dimensional analysis platform according to claim 1, wherein the steps of obtaining the training effect parameter value between the second characteristic distance distribution and the first characteristic distance distribution, updating the to-be-trained anomaly characteristic embedded model waiting for model convergence optimization according to the training effect parameter value, and returning the obtained target training data set to perform iterative learning until the model convergence condition is satisfied, and outputting the converged to-be-trained anomaly characteristic embedded model as a first target anomaly characteristic embedded model, comprise:

6. The method for analyzing the abnormality of the retaining wall structure based on the three-dimensional analysis platform according to claim 1, wherein the target training dataset comprises each multi-modal retaining wall structure data cluster, and the multi-modal retaining wall structure data cluster comprises the associated retaining wall structure data group;

the method further comprises the steps of:

7. The method for analyzing the structural anomaly of a retaining wall based on a three-dimensional analysis platform according to claim 1, wherein after the training effect parameter value between the second characteristic distance distribution and the first characteristic distance distribution is obtained and the to-be-trained anomaly characteristic embedding model waiting for model convergence optimization is updated according to the training effect parameter value, the step of returning the obtained target training data set is performed with iterative learning until the model convergence condition is satisfied, and after outputting the converged to-be-trained anomaly characteristic embedding model as the first target anomaly characteristic embedding model, the method further comprises:

8. The method for analyzing the abnormality of the retaining wall structure based on the three-dimensional analysis platform according to any one of claims 1 to 7, characterized in that the method further comprises:

9. The method for analyzing the retaining wall structure anomaly based on the three-dimensional analysis platform according to claim 8, wherein the obtaining the feature distance between the target anomaly embedded feature to be excavated and the excavated anomaly embedded feature distribution, generating a second feature distance distribution, comprises:

10. A three-dimensional analysis platform-based retaining wall structure anomaly analysis system, comprising a processor and a machine-readable storage medium having stored therein machine-executable instructions loaded and executed by the processor to implement the three-dimensional analysis platform-based retaining wall structure anomaly analysis method of any one of claims 1-9.