A method and system for intelligent identification of slope stability

By using the distributed deployment of cornerstone neural networks and multi-dimensional representation array refinement in slope stability analysis system, the problems of data leakage, high computing power demand and poor model generalization capabilities in slope stability analysis are solved, and the problem of data leakage, high computing power demand and poor model generalization capabilities in slope stability analysis is achieved. Efficient and safe slope stability identification.

CN118799740BActive Publication Date: 2024-11-19CHENGDU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411275464.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-12
Publication Date
2024-11-19
Estimated Expiration
2044-09-12

AI Technical Summary

Technical Problem

The existing slope stability analysis system has the problems of data leakage risk, large network bandwidth consumption, high computing power demand, poor model generalization capabilities, and long training and update cycles, making it difficult to quickly respond to changes in the geological environment.

Method used

The cornerstone neural network is used to divide it into linear mapping components, feature refining components and slope stability recognition components, which are deployed on the local geological analysis platform and the land and resources cloud platform respectively. Through multi-dimensional characterization array refining and iterative adjustment error optimization, it generates Collaborative slope stability identification components to achieve distributed learning and data isolation.

Benefits of technology

It reduces the computing power demand of local geological analysis platforms, improves the speed of tuning and optimization, ensures data security, reduces the number of network interactions, and improves the accuracy and efficiency of slope stability recognition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118799740B_ABST
    Figure CN118799740B_ABST
Patent Text Reader

Abstract

The present application provides a method and system for intelligent identification of slope stability, which performs multi-dimensional characterization array extraction on x first characterization arrays based on a cornerstone neural network to obtain second characterization arrays corresponding to each of the x first characterization arrays, where x≥1; sends one of the x second characterization arrays to x local geological analysis platforms respectively; if the local feature extraction components and local slope stability identification components deployed by the x local geological analysis platforms respectively meet the local network calibration stop requirements, the feature extraction components are optimized for internal configuration parameters based on x target calibration errors to obtain a collaborative feature extraction component; based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms, a collaborative slope stability identification component is generated for performing slope stability analysis on the slope geological data in the target local geological analysis platform.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the fields of data processing and artificial intelligence, and more specifically, to a method and system for intelligently identifying slope stability. Background Art

[0002] In the field of geological disaster prevention and management, slope stability analysis is a vital task. Traditional slope stability analysis methods often rely on expert experience, field surveys, and limited physical model simulations. With the development of information technology, especially the advancement of artificial intelligence technology, more and more research and applications have begun to try to use machine learning and deep learning technologies to automatically identify and evaluate slope stability. These technologies can improve analysis efficiency, reduce labor costs, and with the support of big data, more accurately predict potential risk areas. Existing slope stability analysis systems usually adopt a centralized data processing architecture, and all data collection, processing, and analysis are performed on a central server. Although this architecture can concentrate resources for efficient computing, it also has some problems: a large amount of geological data is uploaded to the central server for processing, which can increase the risk of sensitive data leakage. When the amount of data is huge, frequent data transmission will consume a lot of network bandwidth resources, especially in remote areas where network conditions may be poor. Slope stability analysis usually uses complex machine learning models, which places high demands on the computing power of the central server. The geological environment in different regions varies greatly, and a single global model may be difficult to adapt to all situations, reducing the generalization ability of the model. Due to the large amount of data and computing resources, the training and update cycle of the model is long, making it difficult to respond quickly to new changes in geological conditions. Summary of the invention

[0003] In view of this, the present application provides a method and system for intelligently identifying slope stability.

[0004] According to a first aspect of the present application, a method for intelligent identification of slope stability is provided, which is applied to a land and resources cloud platform, and the method comprises: the land and resources cloud platform obtains x first characterization arrays, and performs multi-dimensional characterization array extraction on the x first characterization arrays based on the feature extraction component in the cornerstone neural network to obtain second characterization arrays corresponding to each of the x first characterization arrays, where x≥1; the cornerstone neural network also comprises a linear mapping component and a slope stability identification component respectively deployed on x local geological analysis platforms; the x first characterization arrays are obtained by extracting features of slope geological data of a monitored target area through the x local geological analysis platforms respectively based on the deployed local feature extraction components; the local feature extraction components deployed by each local geological analysis platform are obtained based on the optimization of internal configuration parameters of the linear mapping component; one of the x second characterization arrays is sent to each of the x local geological analysis platforms; wherein the second characterization array obtained by the target local geological analysis platform corresponds to the first characterization array obtained by feature extraction of the target local geological analysis platform; the target local geological analysis platform is one of the x local geological analysis platforms any local geological analysis platform; if the local feature extraction components and local slope stability identification components respectively deployed by the x local geological analysis platforms meet the local network calibration stop requirements, obtain the target calibration errors respectively sent by the x local geological analysis platforms, wherein the target calibration errors sent by the target local geological analysis platform are constructed by the local slope stability identification components deployed by the target local geological analysis platform and the acquired second characterization array, and the local slope stability identification components deployed by each local geological analysis platform are obtained after optimizing the internal configuration parameters of the slope stability identification components; based on the x target calibration errors, the internal configuration parameters of the feature extraction components are optimized to obtain a collaborative feature extraction component, and based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms, generate the collaborative slope stability identification components corresponding to each local geological analysis platform; at the same time, the collaborative slope stability identification components arranged on the land and resources cloud platform and the target local geological analysis platform are configured to perform slope stability analysis on the slope geological data in the target local geological analysis platform.

[0005] According to a second aspect of the present application, a method for intelligently identifying slope stability is provided, which is applied to a target local geological analysis platform, and the method includes: the target local geological analysis platform linearly maps the slope geological data according to the linear mapping component in the cornerstone neural network to obtain a first characterization array, and sends the first characterization array to the land and resources cloud platform, so that the land and resources cloud platform performs multi-dimensional characterization array extraction on the first characterization array based on the feature extraction component in the cornerstone neural network to obtain a second characterization array; the linear mapping component in the cornerstone neural network is deployed on x local geological analysis platforms, and the cornerstone neural network also includes a slope stability identification component for deployment on x local geological analysis platforms, and the x local geological analysis platforms include the target local geological analysis platform; the second characterization array sent by the land and resources cloud platform is obtained, slope stability identification is performed on the second characterization array based on the slope stability identification component, and an inference identification result is obtained, and an iterative adjustment is obtained according to the prior mark of the slope geological data and the inference identification result. error, optimizing the internal configuration parameters of the linear mapping component and the slope stability identification component based on the iterative calibration error to obtain a local feature extraction component and a local slope stability identification component; if both the local feature extraction component and the local slope stability identification component meet the local network calibration stop requirement, sending the target calibration error generated based on the local slope stability identification component to the land and resources cloud platform, so that the land and resources cloud platform optimizes the internal configuration parameters of the feature extraction component according to the target calibration errors corresponding to each of the x local geological analysis platforms to obtain a collaborative feature extraction component, and generating a collaborative slope stability identification component corresponding to each local geological analysis platform based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms; at the same time, the collaborative slope stability identification component arranged on the land and resources cloud platform and the target local geological analysis platform is configured to perform slope stability analysis on the slope geological data in the target local geological analysis platform.

[0006] According to the third aspect of the present application, there is provided a slope stability intelligent identification system, comprising a land and resources cloud platform and at least one local geological analysis platform communicatively connected to the land and resources cloud platform, wherein the land and resources cloud platform comprises a first processor; and a first memory, wherein a first computer program is stored in the first memory, and when the first computer program is run by the first processor, the first processor executes the method described in the first aspect; the local geological analysis platform comprises a second processor; and a second memory, wherein a second computer program is stored in the second memory, and when the second computer program is run by the second processor, the second processor executes the method described in the second aspect.

[0007] The beneficial effects of the present application include at least: the present application is based on dividing the cornerstone neural network into a linear mapping component, a feature extraction component and a slope stability identification component, and the linear mapping component and the slope stability identification component are sent to each local geological analysis platform respectively, and the land and resources cloud platform maintains the feature extraction component. By adjusting and optimizing the local geological analysis platform and the land and resources cloud platform together, the computing power requirements of the local geological analysis platform can be reduced and the cost can be reduced. The local geological analysis platform and the land and resources cloud platform can be adjusted and optimized at the same time to improve the speed of adjustment and optimization. Each local geological analysis platform linearly maps the slope geological data based on the deployed linear mapping component to obtain a first representation array, and the land and resources cloud platform obtains x first representation arrays, and performs multi-dimensional representation array extraction on the x first representation arrays based on the feature extraction component to obtain the second representation arrays corresponding to each of the x first representation arrays. Send one of the x second characterization arrays to x local geological analysis platforms respectively; the second characterization array obtained by each local geological analysis platform corresponds to the first characterization array obtained by feature extraction of each local geological analysis platform, and each local geological analysis platform can obtain an iterative calibration error based on the deployed slope stability identification component and the obtained second characterization array, and optimize the internal configuration parameters of the linear mapping component and the slope stability identification component based on the iterative calibration error to obtain a local feature extraction component and a local slope stability identification component. If the local feature extraction components and local slope stability identification components deployed by the x local geological analysis platforms meet the local network calibration stop requirements, each local geological analysis platform will use the iterative calibration error generated by the local slope stability identification component that has reached a stable state and the obtained second characterization array as the target calibration error, and the land and resources cloud platform will optimize the internal configuration parameters of the feature extraction component based on the target calibration errors sent by the x local geological analysis platforms to obtain a collaborative feature extraction component. Based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in x local geological analysis platforms, the collaborative slope stability identification components corresponding to each local geological analysis platform are generated. At the same time, the collaborative slope stability identification components set on the land and resources cloud platform and the target local geological analysis platform are configured to perform slope stability analysis on the slope geological data in the target local geological analysis platform. The local geological analysis platform only needs to upload the adjustment error, and the slope geological data does not need to be synchronized. The data is isolated and the security is guaranteed. Because the distribution of slope geological data is non-uniform, the feature extraction component based on the cornerstone neural network with multi-dimensional representation information knowledge can have a stable representation effect on multi-dimensional heterogeneous data on slope geological data without repeatedly adjusting the network.Based on this, when both the local feature extraction component and the local slope stability identification component meet the local network calibration stop requirements, the land and resources cloud platform can obtain the calibration errors corresponding to each local geological analysis platform to complete a parameter optimization. The feature extraction component does not need to repeatedly adjust the internal configuration parameters to prevent network parameter instability, and at the same time reduce the number of interactions between the local platform and the cloud platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A schematic diagram of the intelligent slope stability identification system provided in an embodiment of the present application.

[0009] Figure 2 A schematic diagram of the implementation flow of a method for intelligently identifying slope stability provided in an embodiment of the present application.

[0010] Figure 3 A schematic diagram of the implementation flow of another method for intelligently identifying slope stability provided in an embodiment of the present application.

[0011] Figure 4 A hardware entity schematic diagram of a land and resources cloud platform provided in an embodiment of the present application.

[0012] Figure 5 A schematic diagram of the hardware entity of a local geological analysis platform provided in an embodiment of the present application. DETAILED DESCRIPTION

[0013] The slope stability intelligent identification method provided in the embodiment of the present application can be applied to Figure 1 The slope stability intelligent identification system shown. Among them, the local geological analysis platform 102 communicates with the land and resources cloud platform 104 through the network. The data storage system can store the data that the land and resources cloud platform 104 needs to process. The data storage system can be integrated on the land and resources cloud platform 104, or it can be placed on the cloud or other network servers. Among them, the geological data can be stored in the local storage of the local geological analysis platform 102, and can also be stored in the data storage system or the cloud storage associated with the land and resources cloud platform 104. The local geological analysis platform 102 can be, but is not limited to, various personal computers, laptops, etc. The land and resources cloud platform 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0014] Please refer to Figure 2 , a schematic diagram of the implementation process of a slope stability intelligent identification method applied to a land and resources cloud platform provided in an embodiment of the present application, specifically including steps S100~S400:

[0015] Step S100: The land and resources cloud platform obtains x first characterization arrays, and performs multi-dimensional characterization array extraction on the x first characterization arrays based on the feature extraction component in the cornerstone neural network to obtain second characterization arrays corresponding to each of the x first characterization arrays, where x≥1; the cornerstone neural network also includes a linear mapping component and a slope stability identification component respectively deployed on x local geological analysis platforms; the x first characterization arrays are obtained by extracting features from the slope geological data of the monitored target area through the x local geological analysis platforms respectively based on the deployed local feature extraction components; the local feature extraction components deployed by each local geological analysis platform are obtained based on the optimization of the internal configuration parameters of the linear mapping component.

[0016] In step S100, the land and resources cloud platform obtains x first characterization arrays from x local geological analysis platforms. These local geological analysis platforms are distributed in different geographical areas, and each platform has its own geological monitoring site responsible for collecting local slope geological data. For example, assuming that x is equal to 3, there are three local geological analysis platforms involved, and each platform will collect slope geological data from its respective monitoring site, such as soil moisture, rock type, topography, vegetation coverage and other information. For the sake of ease of explanation, it is assumed here that the data collected by the first local geological analysis platform can be represented as a first characterization array A in the form of a vector. 1 =[a 11 ,a 12 ,a 13 ,…,a 1n ], where a 1i represents the i-th eigenvalue; similarly, the data collected by the second local geological analysis platform can be expressed as A 2 =[a 21 ,a 22 ,a 23 ,…,a 2n ], the data of the third local geological analysis platform can be expressed as A 3 =[a 31 ,a 32 ,a 33 ,…,a 3n ]. These arrays are called first representation arrays.

[0017] Next, the Land and Resources Cloud Platform uses the feature extraction component in the cornerstone neural network to extract multi-dimensional representation arrays from these first representation arrays. The cornerstone neural network is a pre-trained neural network model that already has the ability to extract meaningful features from raw data. The feature extraction component plays a core role in this process. It can learn more abstract and complex representations from low-level features, thereby helping the slope stability identification model to better understand and predict the stability of the slope.

[0018] Specifically, the feature extraction component can use a multi-channel saliency focusing mechanism, such as a multi-head attention mechanism, to capture the correlation between different geological data features. For example, if the interaction between soil moisture and rock type in a certain area is very important for slope stability, the feature extraction component will enhance the correlation between these features through a multi-head attention mechanism. Assume that the first representation array collected by each local geological analysis platform is obtained after multi-channel saliency focusing to obtain the corresponding saliency focusing array. For example, the saliency focusing array of the first local geological analysis platform is represented as B 1 =[b 11 ,b 12 ,b 13 ,…,b 1n ], where b 1i Represents the i-th eigenvalue after saliency focusing processing.

[0019] Then, the land resources cloud platform will fully connect these significant focus arrays and further extract features using a multi-layer perceptron (MLP) to obtain a linear mapping array. For example, the linear mapping array of the first local geological analysis platform is represented by C 1 =[c 11 ,c 12 ,c 13 ,…,c 1n ]. Finally, the land resources cloud platform generates a second representation array based on these linear mapping arrays. For example, the second representation array corresponding to the first local geological analysis platform is represented as D 1 =[d 11 ,d 12 ,d 13 ,…,d 1n ]. These second characterization arrays contain deeper geological feature information, which helps to improve the accuracy of slope stability identification.

[0020] In summary, in step S100, the land and resources cloud platform further refines the first characterization array collected from the local geological analysis platform through the feature extraction component in the cornerstone neural network, and obtains a higher-level second characterization array. This process fully utilizes the advantages of the pre-trained model and enhances the model's ability to identify slope stability.

[0021] Step S200: Send one of the x second characterization arrays to each of the x local geological analysis platforms; wherein the second characterization array acquired by the target local geological analysis platform corresponds to the first characterization array obtained by feature extraction of the target local geological analysis platform; and the target local geological analysis platform is any one of the x local geological analysis platforms.

[0022] In step S100, the land and resources cloud platform has performed multi-dimensional characterization array extraction on the first characterization arrays collected from x local geological analysis platforms through the feature extraction component in the cornerstone neural network to obtain x second characterization arrays. These second characterization arrays contain richer and more abstract geological feature information, which helps to improve the recognition accuracy of slope stability. For example, for the case of x=3, the land and resources cloud platform has processed and obtained three second characterization arrays D1, D2, and D3.

[0023] Next, in step S200, the land and resources cloud platform sends these second characterization arrays back to the corresponding local geological analysis platforms. Specifically, the first local geological analysis platform will receive the second characterization array D1 corresponding to its first characterization array, the second local geological analysis platform will receive D2, and the third local geological analysis platform will receive D3. The "target local geological analysis platform" here refers to any one of the x local geological analysis platforms. Taking the first local geological analysis platform as an example, the platform has extracted features of the slope geological data of the monitored target area through the deployed local feature extraction component before step S100, and obtained a first characterization array A1. After completing the multi-dimensional characterization array refinement, the land and resources cloud platform will send the second characterization array D1 corresponding to A1 to the first local geological analysis platform. After receiving the second characterization array D1, the first local geological analysis platform will use the linear mapping component and slope stability identification component deployed thereon for further processing. The function of the linear mapping component is to map the second characterization array D1 to a feature space that is more suitable for slope stability identification. For example, the linear mapping component can map the second characterization array D1 to a feature space that is more suitable for slope stability identification. 1 =[d 11 ,d 12 ,d 13 ,…,d 1n ] is mapped into a linear combination E 1 =[e 11 ,e 12 ,e 13 ,…,e 1m ], where m may be different from n, indicating the feature dimension after mapping. The calculation process of the linear mapping component can be expressed by a simple linear transformation formula, for example: E 1 =W·D 1 +b; where W is the score array and b is the bias array. These weights and biases are obtained through the training process and are used to optimize the internal configuration parameters.

[0024] Next, the slope stability identification component uses the mapped array E 1Slope stability identification. The slope stability identification component is usually a classifier, which can be a logistic regression model, a support vector machine, or a deep neural network. Taking logistic regression as an example, the output of the classifier can be expressed as: ;in represents the predicted probability value, σ is the sigmoid function, w and b are the weight vector and bias term of the classifier respectively.

[0025] In step S200, the land and resources cloud platform feeds back the second characterization array refined by the multi-dimensional characterization array to the corresponding local geological analysis platforms. These platforms use the second characterization array to optimize their internal configuration parameters, and further improve the accuracy of slope stability identification through linear mapping components and slope stability identification components.

[0026] Step S300: If the local feature extraction components and local slope stability identification components deployed by the x local geological analysis platforms respectively meet the local network calibration stop requirements, obtain the target calibration errors sent by the x local geological analysis platforms respectively, wherein the target calibration error sent by the target local geological analysis platform is constructed by the local slope stability identification component deployed by the target local geological analysis platform and the acquired second characterization array, and the local slope stability identification component deployed by each local geological analysis platform is obtained after optimizing the internal configuration parameters of the slope stability identification component.

[0027] In step S200, the land and resources cloud platform has fed back the second characterization array to each local geological analysis platform. In step S300, the land and resources cloud platform will evaluate whether these platforms have reached the predetermined local network adjustment stop requirements. Here, it is assumed that x=3, that is, there are three local geological analysis platforms participating in this adjustment process. In order to meet the local network adjustment stop requirements, each local geological analysis platform completes a series of specific tasks. First, each local geological analysis platform will be based on the second characterization array D received from the land and resources cloud platform. i (i=1,2,3) to optimize its internal configuration parameters. Taking the first local geological analysis platform as an example, the second characterization array received by the platform is D 1 =[d 11 ,d 12 ,d 13 ,…,d 1n ]. In order to optimize the internal configuration parameters, the platform first uses the linear mapping component deployed on it to map the second characterization array D1 into a feature space suitable for slope stability identification. The mapping process can be expressed as:

[0028] E 1 =W·D 1 +b;

[0029] Among them, E 1 represents the mapped feature vector, W is the score array, and b is the offset array.

[0030] Next, the local geological analysis platform uses the slope stability identification component to perform the slope stability identification task. This component can be a logistic regression model, a support vector machine, or a deep neural network. Taking logistic regression as an example, the output of the slope stability identification component can be expressed as:

[0031] ;

[0032] in represents the predicted probability value, σ is the sigmoid function, w and b are the weight vector and bias term of the classifier respectively.

[0033] In order to optimize the internal configuration parameters, the local geological analysis platform also calculates the loss function. The loss function measures the gap between the predicted result and the actual label. Assuming the actual label is y, the predicted probability is , then the loss function can be in the form of a binary cross entropy loss function, and the calculation formula is as follows:

[0034] ;

[0035] Each local geological analysis platform will calculate its own loss value and send it to the land resources cloud platform as the target calibration error. For example, the target calibration error calculated by the first local geological analysis platform is L1, the target calibration error calculated by the second local geological analysis platform is L2, and the target calibration error calculated by the third local geological analysis platform is L3.

[0036] After receiving these target calibration errors, the land and resources cloud platform will check whether all local geological analysis platforms have met the local network calibration stop requirements. The local network calibration stop requirements may include but are not limited to one of the following conditions: the target calibration error is lower than the preset threshold, the change in the target calibration error after several consecutive iterations is less than a certain threshold, etc. If all local geological analysis platforms have met these requirements, the entire local network calibration process is terminated. In this way, the land and resources cloud platform not only ensures that the local geological analysis platform can effectively optimize the slope stability identification performance, but also ensures that the calibration process of the entire local network is effective and controllable.

[0037] Step S400: Optimize the internal configuration parameters of the feature extraction component based on x target calibration errors to obtain a collaborative feature extraction component; generate a collaborative slope stability identification component corresponding to each local geological analysis platform based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms; at the same time, the collaborative slope stability identification components set on the land and resources cloud platform and the target local geological analysis platform are configured to perform slope stability analysis on the slope geological data in the target local geological analysis platform.

[0038] In step S400, the target calibration errors previously collected from various local geological analysis platforms are used to optimize the process of feature extraction components, and finally a collaborative slope stability identification component is generated. This step ensures that the land and resources cloud platform can integrate information from various places and improve the overall accuracy and reliability of slope stability identification. In step S300, the land and resources cloud platform has collected the target calibration errors of x local geological analysis platforms. In order to optimize the feature extraction component, the land and resources cloud platform will adjust the internal configuration parameters of the feature extraction component based on these target calibration errors. This adjustment process is intended to ensure that the features extracted from the geological data can more accurately reflect the stability state of the slope. Taking three local geological analysis platforms as an example, it is assumed that the land and resources cloud platform has received three target calibration errors, namely L1, L2, and L3.

[0039] First, the land and resources cloud platform will use these target calibration errors to optimize the feature extraction component. Assume that the feature extraction component adopts a convolutional neural network (CNN) architecture, and its internal configuration parameters include convolution kernel size, step size, padding, etc. The land and resources cloud platform can update these parameters through the back propagation algorithm. The core of the back propagation algorithm is to calculate the gradient of the loss function relative to the parameters of each layer in the network, and adjust the parameter value according to the gradient direction. Taking L1 as an example, the land and resources cloud platform will calculate the gradient of the loss function L1 with respect to the parameters of each layer in the feature extraction component, and adjust the parameters based on these gradients. The specific gradient descent formula can be expressed as:

[0040] ;

[0041] Among them, θ j represents the parameters of the jth layer, α is the learning rate, is the loss function L1 with respect to the parameter θ j The partial derivative of .

[0042] After completing the optimization of the feature extraction component, the land and resources cloud platform will build a collaborative feature extraction component, which is a federated network structure. This collaborative feature extraction component will integrate the optimization results of all local geological analysis platforms, so that the features extracted from the slope geological data will be more comprehensive and accurate. Based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components of each local geological analysis platform, the land and resources cloud platform will generate a collaborative slope stability identification component corresponding to each local geological analysis platform. These collaborative slope stability identification components also constitute part of the federated network, which can share the optimized features and model knowledge, but maintain the privacy of their respective data. Taking the first local geological analysis platform as an example, the land and resources cloud platform will generate a collaborative slope stability identification component based on the optimized collaborative feature extraction component and the local feature extraction component and local slope stability identification component of the platform. The collaborative slope stability identification component can be a neural network model containing multiple hidden layers, which is used to process the optimized features and identify the slope stability. The model can use the softmax function as the activation function of the output layer to predict the stability level of the slope. For example, for a neural network with two hidden layers, the calculation of the output layer can be expressed as:

[0043] ;

[0044] Where E 1 is the optimized feature vector of the first local geological analysis platform, W 1 ,W 2 ,W 3 is the score array, b 1 、b 3 is an array of offsets, is the predicted slope stability probability distribution.

[0045] Finally, the Land and Resources Cloud Platform will deploy the collaborative slope stability identification component to itself and various local geological analysis platforms to achieve slope stability analysis of slope geological data. This means that both the Land and Resources Cloud Platform and any local geological analysis platform can use this optimized model to process new slope geological data and conduct accurate slope stability analysis.

[0046] In one implementation, the method provided in this application may further include:

[0047] Step S101: obtaining data element attributes corresponding to each of the x local geological analysis platforms, obtaining multi-dimensional shared data having commonality with the data element attributes, adjusting and optimizing the preset neural network based on the multi-dimensional shared data, and obtaining a cornerstone neural network that reaches a stable state for each dimension;

[0048] Step S102: Send the linear mapping component and slope stability identification component of the cornerstone neural network to x local geological analysis platforms, so that each local geological analysis platform optimizes the internal configuration parameters of the deployed linear mapping component based on the slope geological data of the monitored target area and the feature extraction component deployed on the land and resources cloud platform to obtain the local feature extraction component, and optimizes the internal configuration parameters of the deployed slope stability identification component to obtain the local slope stability identification component.

[0049] In step S101, the land and resources cloud platform obtains the corresponding data element attributes from X local geological analysis platforms. Data element attributes contain a variety of information, such as data types, such as numerical values, text, etc.; the size of the data set; the distribution of data, such as discrete, continuous, etc. This information is crucial for understanding the basic characteristics of the data and selecting appropriate machine learning algorithms. For example, suppose a local geological analysis platform submits rainfall records for a certain area, which is a time series data containing daily rainfall in the past few years. Another platform may provide soil type classification data, which is a category data that includes different soil type labels. Next, the land and resources cloud platform obtains multi-dimensional shared data with commonality based on these data element attributes. These data can be historical slope stability cases, slope change trends under different geological conditions, etc. Then, these multi-dimensional shared data are used to adjust and optimize the preset neural network (such as Transformer).

[0050] Specifically, a simple Transformer architecture can be used as the basis of the cornerstone neural network. Assume that the input data includes multiple features such as rainfall, soil type, and vegetation coverage, and each feature can be encoded into a vector. For example, for the categorical feature of soil type, one-hot encoding can be used to convert it into a vector form, such as soil A is represented as [1,0,0], soil B is represented as [0,1,0], and soil C is represented as [0,0,1]. Next, the land and resources cloud platform uses this data to train the cornerstone neural network until it reaches a stable state, that is, the performance of the model is no longer significantly improved, which means that the model has converged. In this process, a loss function is used to evaluate the difference between the model prediction result and the actual label, and the model parameters are adjusted through the back propagation algorithm to minimize this difference. For example, in order to measure the prediction effect of the model, a cross entropy loss function can be used.

[0051] In step S102, after the cornerstone neural network is constructed, the land resources cloud platform sends the linear mapping component and slope stability identification component in the cornerstone neural network to X local geological analysis platforms. These components contain part of the structure and parameters of the cornerstone neural network for subsequent localized configuration and optimization.

[0052] After receiving these components, each local geological analysis platform will use the slope geological data of the target area monitored by itself and the feature extraction component deployed on the land and resources cloud platform to optimize the internal configuration parameters of the deployed linear mapping component to obtain the local feature extraction component. At the same time, it will also optimize the internal configuration parameters of the deployed slope stability identification component to obtain the local slope stability identification component.

[0053] Suppose a local geological analysis platform monitors a new slope area, and the geological data of this area includes features such as rainfall, soil type, and vegetation coverage. First, the platform will use the linear mapping component in the cornerstone neural network to perform preliminary processing on these features, such as converting categorical features into vector form and normalizing numerical features. Then, through the optimization process, the platform will adjust the parameters in these components to better fit the local data. Taking the slope stability identification component as an example, this component may contain a simple feedforward neural network, which includes an input layer, a hidden layer, and an output layer. The input layer receives the feature vector, the hidden layer is responsible for feature extraction, and the output layer gives the slope stability prediction result. The optimization process involves adjusting the score array and offset array of the hidden layer so that the model can more accurately predict the stability of the slope.

[0054] Assume that the hidden layer has a score array W and an offset array b. Initially, these parameters can be randomly initialized. During the optimization process, these parameters are continuously adjusted by gradient descent to reduce the gap between the predicted results and the actual labels. Specifically, the score array and the offset array can be updated by the following formula:

[0055] ;

[0056] ;

[0057] Among them, α is the learning rate, L is the loss function value, and are the gradients of the loss function with respect to the score array and the bias array, respectively.

[0058] In this way, the Land and Resources Cloud Platform can not only centrally process a large amount of geological data analysis work, but also effectively promote advanced machine learning models to various local geological analysis platforms, improving the accuracy and efficiency of slope stability assessment. This distributed learning method not only reduces the pressure on the central server, but also enables the model to better adapt to the specific conditions of different regions, thereby providing more accurate support for slope stability assessment.

[0059] In one implementation, in step S100, the land and resources cloud platform obtains x first representation arrays, which may specifically include:

[0060] Step S110: obtaining a platform selection ratio, selecting x local geological analysis platforms from y local geological analysis platforms according to the platform selection ratio, sending a calibration command to the x local geological analysis platforms, so that each local geological analysis platform performs linear mapping on the slope geological data based on the local feature extraction component, and obtains a first representation array, wherein 1≤x≤y;

[0061] Step S120: obtaining first representation arrays respectively sent by x local geological analysis platforms.

[0062] In step S110, the land and resources cloud platform determines the platform selection ratio. This ratio determines how many platforms are selected from y local geological analysis platforms to participate in the adjustment process. The selection of the platform selection ratio is based on multiple factors, including but not limited to the historical performance of each platform, data quality, and current resource availability. For example, if y is equal to 20, and the platform selection ratio is set to 0.5, the land and resources cloud platform will select 10 platforms from these 20 platforms for adjustment. Once the number of platforms x to be selected is determined, the land and resources cloud platform will send adjustment commands to these x local geological analysis platforms. These commands instruct the local geological analysis platform to use the previously obtained local feature extraction component to linearly map the slope geological data to generate a first characterization array. These first characterization arrays represent the key features of the slope geological data and are crucial for subsequent slope stability assessments.

[0063] Assume that the land and resources cloud platform selects local geological analysis platform A as one of the participants. After receiving the adjustment command, platform A begins to process the slope geological data it monitors. These data may include multiple features such as rainfall, soil type, vegetation coverage, groundwater level, etc. To simplify the explanation, consider three key features: rainfall r, soil type s, and vegetation coverage v. The local feature extraction component of platform A can be a pre-trained neural network that contains a linear mapping component for converting these features into a set of numerical vectors. Platform A uses the linear mapping component in the local feature extraction component to merge the three feature vectors into a first representation array. This array can be a long vector containing all feature vectors, or a new vector obtained by a specific combination method. For example, if a simple connection method is used, the first representation array f=[r;s;v].

[0064] In step S120, after the local feature extraction component completes its work, the x local geological analysis platforms send their generated first characterization arrays to the land and resources cloud platform. These arrays contain important information about the slope geological data and will be used for subsequent slope stability assessment.

[0065] Continuing with the example of local geological analysis platform A, the first characterization array f generated by the platform contains information on rainfall, soil type, and vegetation coverage. Platform A sends this array to the land resources cloud platform through a secure data transmission channel. The same process will be repeated on the other 9 selected local geological analysis platforms until all selected platforms have completed data transmission.

[0066] After receiving the first characterization arrays sent by all x local geological analysis platforms, the land and resources cloud platform will further process and analyze these arrays. For example, these arrays can be used to train a slope stability identification model. To simplify the explanation, assume that the land and resources cloud platform uses a simple feedforward neural network to train the slope stability identification model. The input of the model is the first characterization array, and the output is the prediction of slope stability. In order to optimize the model, the land and resources cloud platform calculates the difference between the predicted results and the actual labels, and updates the model parameters by gradient descent. In order to measure the prediction effect of the model, the land and resources cloud platform can use mean square error (MSE) as the loss function. In order to optimize the model parameters, the land and resources cloud platform uses gradient descent to update the score array W and the offset array b.

[0067] In one implementation, in step S110, multi-dimensional representation array extraction is performed on the x first representation arrays based on the feature extraction component in the cornerstone neural network to obtain second representation arrays corresponding to each of the x first representation arrays, which may specifically include:

[0068] Step S130: performing multi-channel saliency focusing on the x first representation arrays to obtain saliency focusing arrays corresponding to each of the x first representation arrays;

[0069] Step S140: Perform full connection on the x saliency focus arrays to obtain linear mapping arrays corresponding to each of the x saliency focus arrays, and generate second representation arrays corresponding to each of the x first representation arrays based on the x linear mapping arrays.

[0070] In step S130, the first step of the land and resources cloud platform to perform multi-dimensional representation array extraction on the x first representation arrays based on the feature extraction component in the cornerstone neural network is to perform multi-channel saliency focusing, that is, for each first representation array, the platform will apply a multi-head attention mechanism to highlight those features that are critical to slope stability identification. For example, assume that the land and resources cloud platform has received first representation arrays from 10 local geological analysis platforms. Each first representation array consists of multiple features, such as rainfall, soil type, vegetation coverage, etc. In order to better understand how these features interact and which features are most important for slope stability, the land and resources cloud platform uses a multi-head attention mechanism to perform saliency focusing on these features. Multi-channel saliency focusing allows the platform to focus on information from different representation subspaces at the same time, and each subspace can capture different association patterns. For example, if 4 attention heads are used, each head will produce a specific saliency focus array. These saliency focus arrays reflect the importance of different features and help highlight those features that have an important impact on slope stability.

[0071] Considering the feature vectors in the first representation array, assume that each first representation array contains 10 features, each represented by a real value. For each first representation array, the land and resources cloud platform uses a multi-head attention mechanism to calculate the saliency focus array. Each attention head assigns weights by calculating the similarity between the query, key, and value vectors, which are then used for weighted average calculation. Assuming the dot product attention mechanism is used, for each attention head, the calculation process is as follows:

[0072] The query vector Q, key vector K, and value vector V are all extracted from the first representation array, and each vector has the same length. For example, for the first feature, the query vector can be Q 1 =[q 1 ,q 2 ,…,q 10 ], key vector K 1 =[k 1 ,k 2 ,…,k 10 ], value vector V 1 =[v 1 ,v2 ,…,v 10 ]. The attention score is calculated by dot product attention, the formula is:

[0073] ;

[0074] where d k is the dimension of the key vector.

[0075] After obtaining the saliency focus array, the land and resources cloud platform continues to execute step S140, that is, linearly mapping the saliency focus array using a fully connected layer (multi-layer perceptron MLP) to generate a second representation array. For example, the land and resources cloud platform now has 10 saliency focus arrays, each of which is generated by a multi-head attention mechanism. Next, the platform inputs these arrays into a multi-layer perceptron (MLP) to further refine these features. MLP usually consists of multiple hidden layers, each of which contains several nodes (neurons). The nodes are connected by weights, and nonlinearity is increased by activation functions. In this example, the land and resources cloud platform can use an MLP with two hidden layers, each hidden layer contains 50 nodes, and the activation function can use ReLU (Rectified Linear Unit).

[0076] As an implementation, the calculation process of the MLP may include: the saliency focus array is used as the input layer. For each saliency focus array, the first hidden layer in the MLP will perform a linear transformation on it and then apply an activation function. Assuming that the number of nodes in the hidden layer is 50, the calculation formula for the input vector x after passing through the first hidden layer is h 1 =ReLU(W 1 x+b 1 ), where W 1 and b 1 are score array and offset array respectively. The second hidden layer again performs 1 Perform linear transformation and activation to obtain the final linear mapping array z. The calculation formula is z=ReLU(W 2 h 1 +b 2 ).

[0077] The land and resources cloud platform uses the above process to process each significant focus array to obtain x linear mapping arrays. These linear mapping arrays are then integrated into the second representation array as input for the next step of analysis. In order to train the feature extraction component in the cornerstone neural network, the land and resources cloud platform can use the mean square error (MSE) as the loss function. In order to optimize the model parameters, the land and resources cloud platform uses the gradient descent method to update the score array W and the offset array b. In this way, the land and resources cloud platform can not only centrally process a large amount of geological data analysis work, but also effectively promote advanced machine learning models to various local geological analysis platforms, thereby improving the accuracy and efficiency of slope stability assessment. This distributed learning method not only reduces the pressure on the central server, but also enables the model to better adapt to the specific conditions of different regions, thereby providing more accurate support for slope stability assessment.

[0078] In one implementation, in step S400, optimizing the internal configuration parameters of the feature extraction component based on the x target calibration errors to obtain the collaborative feature extraction component may specifically include:

[0079] Step S410: Based on the x target calibration errors and the internal configuration parameters in the feature extraction component, generate the unit calibration errors corresponding to each of the x target calibration errors, generate a cloud platform bias vector based on the x unit calibration errors and the calibration tuning parameters, optimize the internal configuration parameters of the feature extraction component based on the cloud platform bias vector, and obtain a collaborative feature extraction component.

[0080] In step S410, the first step of optimizing the internal configuration parameters of the feature extraction component based on the x target calibration errors is to generate the unit calibration errors corresponding to each of the x target calibration errors. Next, the land and resources cloud platform generates a cloud platform bias vector based on the x unit calibration errors and the calibration tuning parameters (also known as the network update step size, learning rate), which represents the gradient of the cloud platform network. Finally, the land and resources cloud platform optimizes the internal configuration parameters of the feature extraction component based on the bias vector to obtain a collaborative feature extraction component.

[0081] For example, suppose that the land and resources cloud platform has completed data processing for multiple local geological analysis platforms and obtained x target calibration errors. Each target calibration error is calculated based on the difference between the predicted results and the actual results in the slope stability identification task. In order to further optimize the feature extraction component, the land and resources cloud platform first generates unit versions of these errors. Each local geological analysis platform may return a target calibration error, which reflects the deviation between the slope stability predicted by the model and the actual situation. For example, if there are 10 different slope stability assessment data sets in a certain area, there will be 10 target calibration errors. The land and resources cloud platform calculates the unit version of each target calibration error to ensure that all errors can be compared on a unified scale. This step helps to maintain consistency in the subsequent optimization process.

[0082] Assume that the target calibration error returned by a local geological analysis platform is e=15.0 and the maximum target calibration error is e max =100.0, the minimum target calibration error is e min = 0. The unit calibration error can be calculated by the following formula:

[0083] ;

[0084] The Land and Resources Cloud Platform then generates a Cloud Platform bias vector, which is done based on the x unit calibration error and the calibration tuning parameter. This bias vector indicates the direction of adjustment in the network to reduce the error. The calibration tuning parameter determines the magnitude of the parameter update in each iteration. For example, the Land and Resources Cloud Platform may choose to calibrate the tuning parameter to α = 0.01.

[0085] The land and resources cloud platform uses the gradient descent method to calculate the cloud platform bias vector. The gradient descent method is a commonly used optimization algorithm that updates parameters by calculating the gradient of the loss function with respect to the network parameters. Specifically, the land and resources cloud platform can use the back propagation algorithm to calculate the gradient. Assuming that the land and resources cloud platform is using the mean square error (MSE) as the loss function, the cloud platform bias vector can be expressed as the gradient of the loss function with respect to the network parameters. Let θ represent the network parameters and L represent the loss function, then the cloud platform bias vector It can be calculated by the following formula:

[0086] ;

[0087] The Land and Resources Cloud Platform uses the generated cloud platform bias vector to update the internal configuration parameters in the feature extraction component. This process helps improve the prediction ability of the model and makes it better suitable for slope stability identification tasks. The Land and Resources Cloud Platform uses the gradient descent method to update the internal configuration parameters to minimize the loss function. The formula for updating the internal configuration parameters is:

[0088] ;

[0089] In order to quantify the prediction error of the model, the land and resources cloud platform can use mean square error (MSE) as the loss function.

[0090] The Land and Resources Cloud Platform uses the above process to optimize the feature extraction component so that the model can better extract features related to slope stability. The feature extraction component may contain multiple structures such as convolutional layers and attention mechanisms to capture the complex relationships in slope stability assessment. For example, a convolutional neural network (CNN) with a multi-head attention mechanism can be used to process remote sensing image data. The multi-head attention mechanism can help the model focus on the most important areas in slope stability assessment, such as soil type, terrain characteristics, etc.

[0091] In one implementation, in step S400, based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms, a collaborative slope stability identification component corresponding to each local geological analysis platform is generated, which may specifically include:

[0092] Step S420: performing collaborative integration operations on the internal configuration parameters of the local feature extraction components and the internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms to obtain network adjustment parameters, and sending the network adjustment parameters to the x local geological analysis platforms so that each local geological analysis platform optimizes the parameters of the deployed local feature extraction components and local slope stability identification components according to the network adjustment parameters to obtain collaborative feature extraction components and collaborative slope stability identification components;

[0093] Step S430: If the collaborative feature refining component meets the cloud platform network adjustment stop requirements, the collaborative feature refining component is combined with the collaborative feature extraction component and the collaborative slope stability identification component deployed by each local geological analysis platform to form the collaborative slope stability identification component corresponding to each local geological analysis platform; wherein, the collaborative feature refining component, the collaborative feature extraction component and the collaborative slope stability identification component deployed by the target local geological analysis platform, together form the collaborative slope stability identification component corresponding to the target local geological analysis platform.

[0094] In step S420, the land and resources cloud platform generates a corresponding collaborative slope stability identification component for each local geological analysis platform based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms. This process is divided into two main stages: the collaborative integration operation obtains network adjustment parameters, and sends the network adjustment parameters to the x local geological analysis platforms, so that each local geological analysis platform optimizes the deployed local feature extraction components and local slope stability identification components according to the network adjustment parameters, thereby obtaining collaborative feature extraction components and collaborative slope stability identification components.

[0095] The land and resources cloud platform first performs a collaborative integration operation on the internal configuration parameters of the local feature extraction components and the local slope stability identification components in the x local geological analysis platforms, namely, federated aggregation, which is a distributed learning method that allows local geological analysis platforms to jointly train models without sharing data. The land and resources cloud platform can use convolutional neural networks (CNNs) as the infrastructure for local feature extraction components. These components are designed to extract features from remote sensing images, geological maps, and other geospatial data. The local slope stability identification component can be a deep learning model for predicting slope stability based on information obtained from the local feature extraction component.

[0096] The land and resources cloud platform collects the internal configuration parameters of the local feature extraction components and local slope stability identification components from x local geological analysis platforms and performs federated aggregation. This process aims to find a global optimal solution so that all local geological analysis platforms can benefit from the data of other platforms without directly sharing the data. Internal configuration parameters include, but are not limited to, learning rate, score array, offset array, etc. For example, in a local geological analysis platform, the local feature extraction component may use a convolutional neural network, whose internal configuration parameters include the score array W and the offset array b.

[0097] The land and resources cloud platform aggregates the internal configuration parameters from each local geological analysis platform and calculates the average or other aggregate statistics. For example, if x local geological analysis platforms each have a local feature extraction component, then the average of the score array and offset array of these components can be calculated. The land and resources cloud platform uses a federated aggregation method to calculate network adjustment parameters. Specifically, the land and resources cloud platform calculates the weighted average of the internal configuration parameters of the local feature extraction component and the local slope stability identification component of each local geological analysis platform. Assuming there are x local geological analysis platforms, each platform has a local feature extraction component and a local slope stability identification component, then the network adjustment parameters can be calculated as follows:

[0098] ;

[0099] ;

[0100] Among them, W i and b i are the score array and offset array of the local feature extraction component or the local slope stability identification component of the i-th local geological analysis platform, respectively, and W avg and b avg are the network adjustment parameters of these components.

[0101] For example, the land and resources cloud platform sends the calculated network adjustment parameters to each local geological analysis platform, so that the local geological analysis platform can optimize the parameters of the local feature extraction component and the local slope stability identification component based on these parameters. avg and offset array b avg Sent to each local geological analysis platform. Each local geological analysis platform uses the received network adjustment parameters to update its local feature extraction component and local slope stability identification component. For example, the update formula may be:

[0102] ;

[0103] ;

[0104] Among them, α is the learning rate, W i and b i It is the score array and offset array of the local feature extraction component or the local slope stability identification component of the i-th local geological analysis platform.

[0105] In step S430, if the collaborative feature extraction component meets the cloud platform network calibration stop requirements, the land and resources cloud platform combines the collaborative feature extraction component with the collaborative feature extraction component (a federated network) and the collaborative slope stability identification component (a federated network) deployed by each local geological analysis platform to form the collaborative slope stability identification component corresponding to each local geological analysis platform. The land and resources cloud platform defines a set of network calibration stop requirements to determine when to stop the network calibration process. This is usually based on model performance indicators, such as the loss function value or accuracy on the validation set. For example, when the mean square error (MSE) on the validation set does not decrease significantly for three consecutive iterations, or reaches the preset maximum number of iterations, the land and resources cloud platform believes that the collaborative feature extraction component has met the network calibration stop requirements. The land and resources cloud platform can use the mean square error (MSE) as the loss function on the validation set. When the cloud platform network calibration stop requirements are met, the land and resources cloud platform combines the collaborative feature extraction component with the collaborative feature extraction component and the collaborative slope stability identification component deployed by each local geological analysis platform to form the collaborative slope stability identification component corresponding to each local geological analysis platform.

[0106] The Land and Resources Cloud Platform has optimized the collaborative feature extraction component through the previous process, enabling it to better extract features related to slope stability. The collaborative feature extraction component and the collaborative slope stability identification component, after optimization, can more efficiently extract features from the local data of the local geological analysis platform and predict slope stability.

[0107] In summary, the land and resources cloud platform obtains network adjustment parameters and sends them to x local geological analysis platforms by performing collaborative integration operations, and combines collaborative feature extraction components with components of local geological analysis platforms according to network adjustment stop requirements, thereby achieving effective generation of collaborative slope stability identification components.

[0108] In one implementation, the network adjustment parameters include feature extraction iteration parameters and slope stability identification iteration parameters. Based on this, step S420 performs a collaborative integration operation on the internal configuration parameters of the local feature extraction components and the internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms to obtain the network adjustment parameters, which may specifically include:

[0109] Step S421: obtaining internal configuration parameters of local feature extraction components in x local geological analysis platforms, performing collaborative integration operations on the internal configuration parameters of the x local feature extraction components, and obtaining feature extraction iteration parameters corresponding to each of the x local geological analysis platforms; the feature extraction iteration parameters are used to send to the corresponding local geological analysis platform, and are configured to optimize the parameters of the local feature extraction components deployed on the corresponding local geological analysis platform, and obtain collaborative feature extraction components;

[0110] Step S422: Obtain the internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms, perform collaborative integration operations on the internal configuration parameters of the x local slope stability identification components, and obtain the slope stability identification iteration parameters corresponding to each of the x local geological analysis platforms; the slope stability identification iteration parameters are used to send to the corresponding local geological analysis platform, and are configured to optimize the parameters of the local slope stability identification components deployed on the corresponding local geological analysis platform to obtain a collaborative slope stability identification component.

[0111] In step S421, the land and resources cloud platform performs collaborative integration operations based on the internal configuration parameters of the local feature extraction components in the x local geological analysis platforms to obtain the feature extraction iterative parameters corresponding to each of the x local geological analysis platforms. These feature extraction iterative parameters are used to send to the corresponding local geological analysis platforms and are configured to optimize the parameters of the local feature extraction components deployed on the corresponding local geological analysis platforms to obtain collaborative feature extraction components.

[0112] For example, the land and resources cloud platform first obtains the internal configuration parameters of the local feature extraction component from x local geological analysis platforms. These parameters contain information such as weights and biases learned by the local feature extraction component during training. The local feature extraction component can be based on a convolutional neural network (CNN) architecture, which is used to extract important features related to slope stability from remote sensing images, geological maps, and other geospatial data. The internal configuration parameters include, for example, a score array W and an offset array b. For example, for a local geological analysis platform, the local feature extraction component may use a convolutional neural network, and its internal configuration parameters include multiple layers of score arrays and offset arrays.

[0113] The land and resources cloud platform performs collaborative integration operations on the internal configuration parameter variables of the local feature extraction components of the x local geological analysis platforms. This operation is similar to federated aggregation in federated learning, which allows local geological analysis platforms to jointly train models without sharing data. The land and resources cloud platform aggregates the internal configuration parameter variables of the local feature extraction components from various local geological analysis platforms and calculates the average or other aggregate statistics. For example, if x local geological analysis platforms all have a local feature extraction component, then the average of the score array and offset array of these components can be calculated.

[0114] The Land and Resources Cloud Platform uses a federated aggregation method to calculate the feature extraction iteration parameters. Specifically, the Land and Resources Cloud Platform calculates the weighted average of the internal configuration parameters of the local feature extraction component of each local geological analysis platform. Assuming there are x local geological analysis platforms, and each platform has a local feature extraction component, the feature extraction iteration parameters can be calculated as follows:

[0115] ;

[0116] ;

[0117] in, and are the score array and offset array of the local feature extraction component of the i-th local geological analysis platform, and are the feature extraction iteration parameters of these components.

[0118] The land and resources cloud platform sends the calculated feature extraction iteration parameters to each local geological analysis platform, so that the local geological analysis platform can optimize the parameters of the local feature extraction component based on these parameters. and offset array Sent to each local geological analysis platform. Each local geological analysis platform uses the received feature extraction iteration parameters to update its own local feature extraction component. For example, the update formula is:

[0119] ;

[0120] ;

[0121] Among them, α is the learning rate, and are the score array and offset array of the local feature extraction component of the i-th local geological analysis platform.

[0122] In step S422, the land and resources cloud platform performs a collaborative integration operation based on the internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms, and obtains the slope stability identification iteration parameters corresponding to each of the x local geological analysis platforms. These slope stability identification iteration parameters are used to send to the corresponding local geological analysis platform, and are configured to optimize the parameters of the local slope stability identification components deployed by the corresponding local geological analysis platform to obtain a collaborative slope stability identification component. The land and resources cloud platform obtains the internal configuration parameters of the local slope stability identification components from the x local geological analysis platforms. These parameters include information such as weights and biases learned by the local slope stability identification components during the training process. The local slope stability identification component can be based on a deep learning model, which is used to predict slope stability based on information obtained from the local feature extraction component.

[0123] The land and resources cloud platform will collect the internal configuration parameters of the local slope stability identification components of the x local geological analysis platforms and perform collaborative integration operations. The land and resources cloud platform summarizes the internal configuration parameters of the local slope stability identification components from each local geological analysis platform and calculates the average value or other aggregate statistics. The land and resources cloud platform uses the federal aggregation method to calculate the slope stability identification iteration parameters. Specifically, the land and resources cloud platform calculates the weighted average of the internal configuration parameters of the local slope stability identification components of each local geological analysis platform. Assuming that there are x local geological analysis platforms in total, and each platform has a local slope stability identification component, the slope stability identification iteration parameters can be calculated as follows:

[0124] ;

[0125] ;

[0126] in, and are the score array and offset array of the local slope stability identification component of the \(i\)th local geological analysis platform, and These components are the iterative parameters for slope stability identification.

[0127] The land and resources cloud platform sends the calculated slope stability identification iteration parameters to each local geological analysis platform, so that the local geological analysis platform can optimize the parameters of the local slope stability identification component based on these parameters. and offset array Sent to each local geological analysis platform. Each local geological analysis platform uses the received slope stability identification iteration parameter to update its own local slope stability identification component. For example, the update formula may be:

[0128] ;

[0129] ;

[0130] Among them, α is the learning rate, and are the score array and offset array of the local slope stability identification component of the ith local geological analysis platform.

[0131] To better understand the above calculation process, let's look at a specific example. Assume that the land resources cloud platform collects data from three local geological analysis platforms, each of which has a local feature extraction component and a local slope stability identification component. For the local feature extraction component, the score arrays of each platform are:

[0132] ; ; ;

[0133] The offset arrays are:

[0134]

[0135] The land resources cloud platform calculates the iterative parameters of feature extraction as follows:

[0136] ;

[0137] For the local slope stability identification component, the score arrays for each platform are: ; ; ;

[0138] The offset arrays are: ; ; ;

[0139] The iterative parameters for slope stability identification calculated by the land and resources cloud platform are as follows:

[0140] ;

[0141] The land and resources cloud platform sends the calculated feature extraction iterative parameters and slope stability identification iterative parameters to various local geological analysis platforms so that they can update their local feature extraction components and local slope stability identification components.

[0142] Through the above steps S421 and S422, the land and resources cloud platform effectively obtains the internal configuration parameters of the local feature extraction component and the local slope stability identification component from x local geological analysis platforms, and performs collaborative integration operations to obtain feature extraction iteration parameters and slope stability identification iteration parameters. These iterative parameters are then sent to each local geological analysis platform for them to update their own local feature extraction components and local slope stability identification components. This approach not only improves the performance of the model in the slope stability identification task, but also ensures the security and privacy protection of the data, ensuring that the land and resources cloud platform can provide more accurate and reliable slope stability assessment results. In this way, the land and resources cloud platform can not only centrally process a large amount of geological data analysis work, but also effectively promote advanced machine learning models to various local geological analysis platforms, improving the accuracy and efficiency of slope stability assessment.

[0143] Please refer to Figure 3 The present application also provides a slope stability intelligent identification method from the perspective of a local geological analysis platform. The method is applied to a target local geological analysis platform. Specifically, the method may include the following steps:

[0144] Step S10: the target local geological analysis platform linearly maps the slope geological data according to the linear mapping component in the cornerstone neural network to obtain a first characterization array, and sends the first characterization array to the land and resources cloud platform, so that the land and resources cloud platform performs multi-dimensional characterization array extraction on the first characterization array based on the feature extraction component in the cornerstone neural network to obtain a second characterization array; the linear mapping component in the cornerstone neural network is deployed on x local geological analysis platforms, the cornerstone neural network also includes a slope stability identification component for deployment on the x local geological analysis platforms, and the x local geological analysis platforms include the target local geological analysis platform;

[0145] Step S20: Obtain the second characterization array sent by the land and resources cloud platform, perform slope stability identification on the second characterization array based on the slope stability identification component, obtain the inference identification result, obtain the iterative adjustment error based on the priori mark of the slope geological data and the inference identification result, optimize the internal configuration parameters of the linear mapping component and the slope stability identification component based on the iterative adjustment error, and obtain the local feature extraction component and the local slope stability identification component;

[0146] Step S30: If both the local feature extraction component and the local slope stability identification component meet the local network calibration stop requirements, the target calibration error generated based on the local slope stability identification component is sent to the land and resources cloud platform, so that the land and resources cloud platform optimizes the internal configuration parameters of the feature extraction component according to the target calibration errors corresponding to each of the x local geological analysis platforms, and obtains a collaborative feature extraction component. Based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms, a collaborative slope stability identification component corresponding to each local geological analysis platform is generated; at the same time, the collaborative slope stability identification component set on the land and resources cloud platform and the target local geological analysis platform is configured to perform slope stability analysis on the slope geological data in the target local geological analysis platform.

[0147] In step S10, the local geological analysis platform linearly maps the slope geological data according to the linear mapping component in the cornerstone neural network to obtain a first representation array, and sends the array to the land and resources cloud platform. The land and resources cloud platform performs multi-dimensional representation array extraction on the first representation array based on the feature extraction component in the cornerstone neural network to obtain a second representation array.

[0148] For example, the linear mapping component in the cornerstone neural network deployed by the local geological analysis platform is mainly used to extract key features from slope geological data. These features may include terrain slope, soil type, groundwater level, etc., which are critical to slope stability. Slope geological data is, for example, data obtained from sources such as satellite images, photos taken by drones, ground radar scans, geological survey reports, etc. The linear mapping component, for example, consists of one or more fully connected layers, which are used to convert the input data into an intermediate representation array, namely the first representation array. Assuming that the local geological analysis platform receives a high-resolution satellite image containing a slope, the linear mapping component converts this image into a vector that contains the key features in the image.

[0149] First representation array: The linear mapping component of the local geological analysis platform converts the input slope image into a vector of length 1000, denoted as v 1 The local geological analysis platform will obtain the first characterization array v 1 The local geological analysis platform sends the v 1 Send to the land and resources cloud platform.

[0150] The land and resources cloud platform uses the feature extraction component to analyze the first characterization array v sent by the local geological analysis platform 1Perform multi-dimensional representation array extraction to obtain the second representation array v 2 The feature extraction component may include multiple fully connected layers and non-linear activation functions to extract higher-level features from the first representation array.

[0151] The feature extraction component of the land and resources cloud platform includes two fully connected layers. The output dimension of the first fully connected layer is 500, and the output dimension of the second fully connected layer is 100. The land and resources cloud platform uses the feature extraction component to convert v 1 Convert to a vector v of length 100 2 .

[0152] In step S20, the local geological analysis platform obtains the second representation array v sent by the land and resources cloud platform 2 , based on the deployed slope stability identification component, the slope stability identification of the second characterization array is performed to obtain the inference identification result. According to the priori mark of the slope geological data and the inference identification result, the iterative adjustment error is obtained, and the internal configuration parameter of the linear mapping component and the slope stability identification component is optimized accordingly to obtain the local feature extraction component and the local slope stability identification component.

[0153] For example, the local geological analysis platform uses the deployed slope stability identification component to characterize the second array v 2 Slope stability identification is performed. The slope stability identification component can be a multi-layer perceptron (MLP) to predict whether the slope is stable. Assume that the slope stability identification component deployed by the local geological analysis platform contains two fully connected layers, the output dimension of the first fully connected layer is 50, and the output dimension of the second fully connected layer is 1 (indicating the slope stability prediction value). The local geological analysis platform uses the slope stability identification component to identify v 2 A prediction is made and a probability value between 0 and 1 is obtained, which represents the probability of slope stability.

[0154] In order to quantify the difference between the predicted results of the slope stability identification component and the actual label, the local geological analysis platform calculates the iterative calibration error. Assume that the prior label of the slope geological data is a binary label (0 for unstable and 1 for stable). In the inference identification result, it is assumed that the probability value predicted by the slope stability identification component is 0.7. The local geological analysis platform can use binary cross entropy loss as the calculation standard for the iterative calibration error. The local geological analysis platform uses the iterative calibration error to optimize the internal configuration parameters of the linear mapping component and the slope stability identification component to improve the prediction performance of the model. The local geological analysis platform can use the gradient descent algorithm to update the internal configuration parameters of the linear mapping component and the slope stability identification component.

[0155] After parameter optimization, the local geological analysis platform obtained the local feature extraction component and the local slope stability identification component. These components will be used for subsequent slope stability analysis tasks. The local feature extraction component is composed of the optimized linear mapping component, and the local slope stability identification component is composed of the optimized slope stability identification component.

[0156] In step S30, if both the local feature extraction component and the local slope stability identification component have reached the local network calibration stop requirement, the local geological analysis platform will send the target calibration error generated based on the local slope stability identification component to the land and resources cloud platform. The land and resources cloud platform optimizes the internal configuration parameters of the feature extraction component based on the target calibration errors corresponding to each of the x local geological analysis platforms to obtain a collaborative feature extraction component. Based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms, a collaborative slope stability identification component corresponding to each local geological analysis platform is generated. The collaborative slope stability identification component will be set on the land and resources cloud platform and the target local geological analysis platform to perform slope stability analysis on the slope geological data in the target local geological analysis platform.

[0157] For example, the local geological analysis platform stops training the local model when a certain convergence condition is reached. This usually means that the performance of the model on the validation set no longer improves significantly. Regarding the local network adjustment stopping requirement, for example, when the performance of the local slope stability identification component on the validation set does not improve significantly within 5 consecutive iterations, the local geological analysis platform considers that the model has converged.

[0158] The local geological analysis platform sends the target calibration error generated by the local slope stability identification component to the land and resources cloud platform. This helps the land and resources cloud platform understand the model performance of each local geological analysis platform.

[0159] Based on the collaborative feature refining component and the local feature extraction components and local slope stability identification components in x local geological analysis platforms, the land and resources cloud platform generates a corresponding collaborative slope stability identification component for each local geological analysis platform. The collaborative slope stability identification component is, for example, an enhanced slope stability identification model generated by the land and resources cloud platform by combining the collaborative feature refining component with the local feature extraction component and the local slope stability identification component of the local geological analysis platform. The collaborative slope stability identification component is set on the land and resources cloud platform and the target local geological analysis platform to perform slope stability analysis on the slope geological data in the target local geological analysis platform. The local geological analysis platform can use the collaborative slope stability identification component deployed locally to perform real-time analysis on new slope geological data to predict the stability of the slope. At the same time, the land and resources cloud platform can also use the collaborative slope stability identification component to provide services to other local geological analysis platforms.

[0160] Through the above steps S10, S20 and S30, the local geological analysis platform effectively extracts key features from the slope geological data and uses the linear mapping component and slope stability identification component in the cornerstone neural network to identify slope stability. The cooperation between the local geological analysis platform and the land and resources cloud platform not only improves the performance of the model in the slope stability identification task, but also ensures the security and privacy protection of the data, ensuring that the local geological analysis platform can provide more accurate and reliable slope stability assessment results.

[0161] In one embodiment, in step S10, the target local geological analysis platform performs linear mapping on the slope geological data according to the linear mapping component in the cornerstone neural network to obtain a first representation array, which may specifically include:

[0162] Step S11: Decompose the slope geological data into i geological data items, perform linear mapping on the i geological data items respectively, and obtain data item representation arrays corresponding to the i geological data items, i≥1;

[0163] Step S12: embed the distribution information of each geological data item in the slope geological data to obtain i distribution representation arrays, fuse the i data item representation arrays and the i distribution representation arrays to obtain a transition representation array, fuse the transition representation array and the identification coding array to obtain a first representation array, wherein the identification coding array covers the overall feature information of the i data item representation arrays.

[0164] In step S11, the local geological analysis platform decomposes the slope geological data into i geological data items according to the linear mapping component in the cornerstone neural network, and linearly maps these geological data items respectively to obtain the data item representation array corresponding to each of the i geological data items.

[0165] For example, the local geological analysis platform first decomposes the slope geological data into i geological data items for subsequent processing. These geological data items may include terrain slope maps, soil type maps, groundwater level maps, etc. Assume that the local geological analysis platform obtains a set of slope geological data including terrain slope, soil type and groundwater level from various sensors and remote sensing equipment. The local geological analysis platform decomposes this set of data into three geological data items: terrain slope map, soil type map and groundwater level map, that is, i=3. For each geological data item, the local geological analysis platform uses a linear mapping component to convert it into a data item representation array. The linear mapping component can be a simple linear transformation, such as matrix multiplication, or a complex neural network structure, such as a convolutional neural network (CNN). For example, for the terrain slope map, the local geological analysis platform can use a convolution layer to extract key features. After each geological data item passes through the linear mapping component, it is converted into a fixed-length vector, called a data item representation array. Assume that the terrain slope map is converted into a vector v of length 100 after the linear mapping component. 1 , the soil type map is converted into a vector v of length 100 2 , the groundwater level map is converted into a vector v of length 100 3 .

[0166] In step S12, the local geological analysis platform embeds the distribution information of each geological data item in the slope geological data to obtain i distribution representation arrays. Next, the i data item representation arrays and the i distribution representation arrays are fused (such as splicing) to obtain a transition representation array. Finally, the transition representation array and the identification coding array are fused to obtain a first representation array. For example, in order to better capture the relationship between geological data items, the local geological analysis platform embeds the distribution information of each geological data item in its slope geological data to obtain i distribution representation arrays. Assume that the local geological analysis platform uses an additional neural network layer to embed distribution information. For example, for a terrain slope map, the local geological analysis platform can use a network containing multiple fully connected layers to capture the distribution pattern of the terrain slope in the entire slope geological data. After each geological data item is embedded with distribution information, it is converted into a distribution representation array. Assume that the terrain slope map is converted into a vector d of length 50 after distribution information embedding. 1 , the soil type map is converted into a vector d of length 50 2 , the groundwater level map is converted into a vector d of length 50 3 .

[0167] The local geological analysis platform performs a fusion operation (such as splicing) on ​​i data item representation arrays and i distribution representation arrays to obtain a transition representation array. Assume that the local geological analysis platform uses a simple vector splicing method to fuse the data item representation array and the distribution representation array. For example, the data item representation array v of the terrain slope map 1 and distribution representation array d 1 After concatenation, a vector t with a length of 150 is formed 1 Similarly, the soil type map and groundwater level map are processed in the same way to form t 2 and t 3 After the three geological data items are fused, three transition representation arrays are formed: 1 ,t 2 and t 3 .

[0168] The local geological analysis platform obtains the identification coding array in some way, which covers the overall characteristic information of the i-item representation array. The identification coding array can be obtained in a variety of ways. A common way is to use an attention mechanism to aggregate the information of all data item representation arrays. Another way is to generate the array through an independent neural network layer. For example, suppose that the local geological analysis platform uses an attention mechanism to obtain the identification coding array. Specifically, a fully connected layer with a softmax activation function can be used to calculate the importance weight of each geological data item, and then the data item representation array is weighted and summed according to these weights to form an identification coding array. Assume that the length of the identification coding array is 100, denoted by c.

[0169] The local geological analysis platform performs a fusion operation on the transition characterization array and the identification code array to obtain a first characterization array. Assume that the local geological analysis platform also uses a simple vector splicing method to fuse the transition characterization array and the identification code array. For example, three transition characterization arrays t 1 ,t 2 and t 3 And the recognition code array c is concatenated to form a vector f with a length of 550. The final vector f is the first representation array, and its length is 550.

[0170] Through the above steps S11 and S12, the local geological analysis platform effectively extracts key features from the slope geological data and uses the linear mapping component in the cornerstone neural network to perform preliminary feature extraction. By disassembling the geological data items, linear mapping, embedding the distribution information, and fusing the representation array, the local geological analysis platform can obtain a first representation array that integrates the characteristics and distribution information of the geological data items, laying the foundation for further slope stability analysis.

[0171] In one implementation, in step S20, slope stability identification is performed on the second characterization array based on the slope stability identification component to obtain an inference identification result, which may specifically include:

[0172] Step S21: solving the inner product of the second representation array and the score array to obtain a weight representation array, where the score array represents the mapping relationship between the second representation array and the priori label set;

[0173] Step S22: Fusing the weight characterization array and the offset array to obtain a projection array, standardizing the projection array, obtaining the reasoning confidence corresponding to each slope stability prior marker in the prior marker set, and using the slope stability prior marker corresponding to the maximum reasoning confidence in the prior marker set as the reasoning recognition result of the second characterization array, wherein the offset array is an adaptively adjusted parameter array.

[0174] In step S21, the local geological analysis platform performs slope stability identification on the second characterization array based on the slope stability identification component. This process first involves solving the inner product with the score array to obtain the weighted characterization array. The score array represents the mapping relationship between the second characterization array and the priori tag set.

[0175] The local geological analysis platform uses a score array to characterize the mapping relationship between the second characterization array and the prior mark set. Assume that the prior mark set contains different categories of slope stability marks, such as stable, slightly unstable, severely unstable, etc. The score array is a two-dimensional array, that is, a weight matrix, which defines the correlation between each slope stability prior mark and the second characterization array. Assume that the score array is W, which is an m×n matrix, where m is the number of categories in the prior mark set and n is the length of the second characterization array. Assume that the prior mark set contains three categories: stable, slightly unstable, and severely unstable, and the corresponding category numbers are 1, 2, and 3 respectively. The length of the second characterization array is 500. Therefore, the shape of the score array W is 3×500.

[0176] The local geological analysis platform solves the inner product of the second characterization array and the score array to obtain the weighted characterization array. The local geological analysis platform obtains the weighted characterization array by calculating the inner product between the second characterization array and the score array. Assuming that the second characterization array is z and its length is 500, the calculation formula of the weighted characterization array w is: w=W·z; where · represents matrix multiplication operation.

[0177] In step S22, the local geological analysis platform fuses the weight representation array and the offset array to obtain a projection array. Then, the projection array is standardized to obtain the reasoning confidence corresponding to each slope stability priori mark in the priori mark set. Finally, the slope stability priori mark corresponding to the maximum reasoning confidence in the priori mark set is used as the reasoning recognition result of the second representation array.

[0178] The local geological analysis platform uses an offset array to adjust the weight representation array to better match the actual slope stability. The offset array is an adaptively adjusted parameter array, that is, a learnable parameter vector, which is used to adjust the weight representation array. Assume that the offset array is b, and its length is equal to the number of categories in the prior tag set. Assume that the prior tag set contains three categories, so the length of the offset array b is 3.

[0179] The local geological analysis platform performs a fusion operation on the weight representation array and the offset array to obtain a projection array. The local geological analysis platform obtains the projection array by adding the weight representation array and the offset array. Assuming that the weight representation array is w and the offset array is b, the calculation formula of the projection array p is: P=w+b;

[0180] The local geological analysis platform normalizes the projection array to obtain the corresponding reasoning confidence of each slope stability prior marker in the prior marker set. The local geological analysis platform usually uses the softmax function to normalize the projection array so that each element becomes a probability form and the sum of all elements is 1. Assuming that the projection array is p and the normalized array is s, the normalization formula is: ; Among them, p i is the i-th element in the projection array, and exp represents the natural exponential function.

[0181] The local geological analysis platform selects the slope stability prior marker corresponding to the maximum reasoning confidence in the prior marker set as the reasoning recognition result of the second characterization array. The local geological analysis platform determines the reasoning recognition result based on the maximum value in the standardized array s. Assuming that the index corresponding to the maximum value in the standardized array s is j, the reasoning recognition result is the jth marker in the prior marker set. Assuming that the values ​​of the standardized array s are [0.1, 0.7, 0.2], the index corresponding to the maximum value 0.7 is 1, so the reasoning recognition result is the second marker in the prior marker set - slightly unstable.

[0182] In order to quantify the process of slope stability identification, some specific calculation formula examples are given here.

[0183] Assume that the local geological analysis platform uses the score array w and the second representation array z to solve the inner product to obtain the weight representation array w. Assume that the shape of the score array w is 3×500, that is, it contains the weights of three categories (stable, slightly unstable, and severely unstable). Assume that the length of the second representation array z is 500. The calculation formula of the weight representation array w is: w=W·z; where · represents matrix multiplication operation.

[0184] Assume that the local geological analysis platform uses the offset array b to fuse with the weight representation array w to obtain the projection array p. Assume that the length of the offset array b is 3, which is the same as the number of categories in the prior tag set. Assume that the local geological analysis platform uses the softmax function to standardize the projection array p to obtain the inference confidence s corresponding to each slope stability prior tag in the prior tag set.

[0185] Assume that the local geological analysis platform determines the inference recognition result based on the maximum value in the standardized array s. Assume that the values ​​of the standardized array s are [0.1, 0.7, 0.2], then the inference recognition result is the second mark in the prior mark set - slightly unstable.

[0186] Through the above steps S21 and S22, the local geological analysis platform effectively identified the slope stability of the second characterization array and obtained the reasoning identification result. This process involves solving the inner product with the score array to obtain the weighted characterization array, and then fusing and normalizing with the offset array to obtain the reasoning confidence, and finally determining the reasoning identification result of the slope stability prior mark.

[0187] In one embodiment, in step S20, optimizing the internal configuration parameters of the linear mapping component and the slope stability identification component based on the iterative calibration error to obtain the local feature extraction component and the local slope stability identification component may specifically include:

[0188] Step S23: Generate feature extraction error parameters according to the iterative calibration error and the internal configuration parameters of the linear mapping component, solve the inner product of the feature extraction error parameters and the calibration tuning parameters to obtain feature extraction iterative bias vectors, optimize the internal configuration parameters of the linear mapping component according to the feature extraction iterative bias vectors, and obtain the local feature extraction component;

[0189] Step S24: Generate identification error parameters based on the iterative calibration error and the internal configuration parameters of the slope stability identification component, solve the inner product of the identification error parameters and the calibration tuning parameters to obtain the identification iterative deviation vector, optimize the internal configuration parameters of the slope stability identification component based on the identification iterative deviation vector, and obtain the local slope stability identification component.

[0190] In step S23, the local geological analysis platform generates a feature extraction error parameter based on the iterative calibration error and the internal configuration parameters of the linear mapping component. Next, the local geological analysis platform solves the inner product of the feature extraction error parameter and the calibration tuning parameter to obtain the feature extraction iterative bias vector (i.e., iterative gradient). The internal configuration parameters of the linear mapping component are optimized based on the feature extraction iterative bias vector to obtain the local feature extraction component. The local geological analysis platform uses the iterative calibration error to measure the gap between the predicted slope stability and the actual slope stability. Assuming that the local geological analysis platform uses the mean square error as the calculation method for the iterative calibration error, the calculation formula is: ; where E represents the iterative calibration error, N is the number of samples, and y n and represent the actual slope stability and predicted slope stability of the nth sample respectively.

[0191] The local geological analysis platform uses a linear mapping component to transform the input features into a space that is more conducive to slope stability identification. Assume that the linear mapping component consists of a score array w and an offset array b. The score array w defines the linear relationship between the input feature vector and the output feature vector. Assuming that the dimension of the input feature vector is d and the dimension of the output feature vector is h, then the shape of the score array is h×d. The offset array b is used to adjust the baseline value of the output feature vector. Assuming that the dimension of the output feature vector is h, then the shape of the offset array is h×1.

[0192] The local geological analysis platform solves the inner product of the feature extraction error parameter and the adjustment tuning parameter to obtain the feature extraction iterative bias vector (i.e., iterative gradient). f Describes the error in the feature extraction process. Assume that e f The shape of is the same as the score array w, both are h×d. The tuning parameter t is used to adjust the direction and size of the iterative gradient. Assume that the shape of t is h×d. Feature extraction iterative bias vector g f Used to update the internal configuration parameters of the linear mapping component. Assume g f The calculation formula is: f =e f ·t. The local geological analysis platform optimizes the internal configuration parameters of the linear mapping component based on the feature extraction iterative bias vector to obtain the local feature extraction component. Assume that the local geological analysis platform uses the gradient descent method to update the score array w and the offset array b. Assuming that the learning rate is α, the update formula of the score array w is: W:=W-α·g f ; The update formula of the offset array b is: b:=b-α·g b ; Among them, g b Forf Similar to the iterative gradient of the bias term.

[0193] In step S24, the local geological analysis platform generates an identification error parameter based on the iterative calibration error and the internal configuration parameter of the slope stability identification component. Next, the local geological analysis platform solves the inner product of the identification error parameter and the calibration tuning parameter to obtain the identification iterative deviation vector. The internal configuration parameter of the slope stability identification component is optimized based on the identification iterative deviation vector to obtain the local slope stability identification component.

[0194] The local geological analysis platform uses the slope stability identification component to predict the stability of the slope. Assume that the slope stability identification component consists of a score matrix V and an offset array c. The score matrix V defines the degree of association between the eigenvector and the different slope stability categories. Assuming that the dimension of the output eigenvector is h and the number of slope stability categories is k, the shape of the score matrix is ​​k×h. The offset array c is used to adjust the prediction probability of different slope stability categories. Assuming that the number of slope stability categories is k, the shape of the offset array is k×1.

[0195] The local geological analysis platform solves the inner product of the identification error parameter and the adjustment tuning parameter to obtain the identification iterative bias vector. r Describes the error in the slope stability identification process. Assume that e r The shape of is the same as the score matrix V, both are k×h. Identify the iterative bias vector g r Used to update the internal configuration parameters of the slope stability identification component. Assume that g r The calculation formula is: r =e r ·t.

[0196] The local geological analysis platform optimizes the internal configuration parameters of the slope stability identification component based on the identification iterative deflection vector to obtain the local slope stability identification component. Assume that the local geological analysis platform also uses the gradient descent method to update the score matrix V and the offset array c. Assuming that the learning rate is α, the update formula of the score matrix V is: V:=V-α·g r ; The update formula of the offset array c is: c:=c-α·g c ; Among them, g c For r Similar to the iterative gradient of the bias term.

[0197] Assume that the local geological analysis platform is processing a slope stability dataset that contains 1,000 samples, each with 100 features. The goal of the local geological analysis platform is to accurately identify whether the slope is stable by training the model.

[0198] The local geological analysis platform calculates the iterative adjustment error. Assume that the local geological analysis platform uses the mean square error as the calculation method for iterative adjustment error. Given the predicted slope stability and actual slope stability y n , the calculation formula of iterative adjustment error E is:

[0199] ;

[0200] Feature extraction iterative bias vector g f The calculation formula is: f =e f t;

[0201] The local geological analysis platform extracts the iterative deviation vector g based on the characteristics f Update the internal configuration parameters of the linear mapping component. Assume that the learning rate is α=0.01.

[0202] The update formula of the score array w is:

[0203] ;

[0204] The update formula of the offset array b is:

[0205] ;

[0206] Identify the iterative deflection vector g r The calculation formula is:

[0207] ;

[0208] The local geological analysis platform is based on the identification of iterative deflection vector g r Update the internal configuration parameters of the slope stability identification component. Assume that the learning rate is α=0.01.

[0209] The update formula of the score matrix V is:

[0210] ;

[0211] The update formula of the offset array c is:

[0212] ;

[0213] Through the above steps S23 and S24, the local geological analysis platform effectively optimizes the internal configuration parameters of the linear mapping component and the slope stability identification component, and obtains the local feature extraction component and the local slope stability identification component. This process involves generating feature extraction error parameters and identification error parameters, and solving the inner product with the adjustment and tuning parameters to obtain iterative gradients, and finally updating the internal configuration parameters of the components based on these iterative gradients.

[0214] In one embodiment, the above method may further include:

[0215] Step S301: obtaining network adjustment parameters sent by the land and resources cloud platform; the network adjustment parameters include feature extraction iteration parameters and slope stability identification iteration parameters, the feature extraction iteration parameters are obtained by the land and resources cloud platform through collaborative integration of internal configuration parameters of local feature extraction components in x local geological analysis platforms, and the slope stability identification iteration parameters are obtained by the land and resources cloud platform through collaborative integration of internal configuration parameters of local slope stability identification components in x local geological analysis platforms;

[0216] Step S302: performing inner product calculation on the feature extraction iteration parameter and the first collaborative training iteration weight to obtain a collaborative feature extraction bias vector, optimizing the parameters of the local feature extraction component deployed on the target local geological analysis platform according to the collaborative feature extraction bias vector, and obtaining the collaborative feature extraction component corresponding to the target local geological analysis platform;

[0217] Step S303: Solve the inner product of the feature extraction iteration parameter and the second collaborative training iteration weight to obtain the collaborative identification bias vector, and optimize the parameters of the local slope stability identification component deployed on the target local geological analysis platform based on the collaborative feature extraction bias vector to obtain the collaborative slope stability identification component corresponding to the target local geological analysis platform.

[0218] In step S301, the local geological analysis platform obtains network adjustment parameters sent by the land and resources cloud platform, including feature extraction iteration parameters and slope stability identification iteration parameters. These parameters are obtained by the land and resources cloud platform through the coordinated integration of the internal configuration parameters of the local feature extraction components and local slope stability identification components in multiple local geological analysis platforms.

[0219] The land and resources cloud platform collects the internal configuration parameters of the local feature extraction components and local slope stability identification components from multiple local geological analysis platforms. For each local geological analysis platform, these parameters include score arrays and offset arrays. The land and resources cloud platform integrates these parameters by averaging or weighted averaging to generate feature extraction iteration parameters and slope stability identification iteration parameters. Assume that the land and resources cloud platform has collected the internal configuration parameters of the local feature extraction components from five local geological analysis platforms. These local geological analysis platforms each have their own specific score array W i and offset array b i The land resources cloud platform can generate feature extraction iterative parameters by simple averaging or weighted averaging. and .

[0220] ;

[0221] ;

[0222] Similarly, the Land and Resources Cloud Platform collected the internal configuration parameters of the local slope stability identification component from five local geological analysis platforms. These local geological analysis platforms each have their own specific score matrix V i and offset array c i The Land and Resources Cloud Platform can generate slope stability identification iterative parameters by simple averaging or weighted averaging. and .

[0223] ;

[0224] ;

[0225] In step S302, the local geological analysis platform performs inner product calculation on the feature extraction iteration parameter and the first collaborative training iteration weight to obtain a collaborative feature extraction bias vector. The local feature extraction component deployed by the target local geological analysis platform is optimized according to the collaborative feature extraction bias vector to obtain a collaborative feature extraction component corresponding to the target local geological analysis platform.

[0226] The local geological analysis platform receives the feature extraction iterative parameters sent by the land and resources cloud platform and At the same time, the local geological analysis platform also has a set of first collaborative training iteration weights t fe and u fe , used to adjust the direction and size of the iterative gradient.

[0227] Collaborative feature extraction bias vector g feUsed to update the internal configuration parameters of the local feature extraction component. Assume g fe The calculation formula is:

[0228] ;

[0229] The local geological analysis platform extracts the deviation vector g based on the collaborative feature fe The parameters of the local feature extraction component are optimized to obtain the collaborative feature extraction component.

[0230] Assume that the local geological analysis platform uses the gradient descent method to update the score array w and the offset array b. Assuming the learning rate is α, the update formula of the score array w is:

[0231] ;

[0232] The update formula of the offset array b is:

[0233] ;

[0234] Among them, g b For fe Similar to the iterative gradient of the bias term.

[0235] In step S303, the local geological analysis platform performs inner product calculation on the slope stability identification iteration parameter and the second collaborative training iteration weight to obtain the collaborative identification deviation vector. The local slope stability identification component deployed by the target local geological analysis platform is optimized according to the collaborative identification deviation vector to obtain the collaborative slope stability identification component corresponding to the target local geological analysis platform.

[0236] The local geological analysis platform receives the slope stability identification iterative parameters sent by the land and resources cloud platform and At the same time, the local geological analysis platform also has a set of second collaborative training iteration weights t r and u r , used to adjust the direction and size of the iterative gradient.

[0237] Collaborative identification of deflection vector g r Used to update the internal configuration parameters of the local slope stability identification component. Assume that g r The calculation formula is:

[0238] ;

[0239] Local geological analysis platform based on collaborative identification of deviation vector g r The parameters of the local slope stability identification component are optimized to obtain the coordinated slope stability identification component.

[0240] Assume that the local geological analysis platform uses the gradient descent method to update the score matrix V and the offset array c. Assuming the learning rate is α, the update formula of the score matrix V is:

[0241] ;

[0242] The update formula of the offset array c is:

[0243] ;

[0244] Among them, g c For r Similar to the iterative gradient of the bias term.

[0245] Assume that the local geological analysis platform is processing a set of slope stability data sets, which contains 1000 samples, each with 100 features. The goal of the local geological analysis platform is to accurately identify whether the slope is stable by training the model. Input feature vector: The feature vector length of each sample is 100, that is, (d=100). Output feature vector: After the linear mapping component, the length of the output feature vector is 50, that is, (h=50). Slope stability category: Slope stability is divided into three categories: stable, slightly unstable, and severely unstable, that is, (k=3).

[0246] The Land and Resources Cloud Platform collects the internal configuration parameters of the local feature extraction components and local slope stability identification components from five local geological analysis platforms. These local geological analysis platforms each have their own specific score arrays and offset arrays, as well as score matrices and offset arrays. The Land and Resources Cloud Platform generates feature extraction iteration parameters and slope stability identification iteration parameters by averaging or weighted averaging these parameters.

[0247] Assume that the score arrays and offset arrays of the five local geological analysis platforms are and . Feature extraction iterative parameters generated by the land and resources cloud platform and As shown below:

[0248] ;

[0249] ;

[0250] Collaborative feature extraction bias vector g fe The calculation formula is:

[0251] ;

[0252] Through the above steps S301 to S303, the local geological analysis platform effectively utilizes the network adjustment parameters provided by the land and resources cloud platform, obtains the collaborative feature extraction bias vector and the collaborative identification bias vector through the inner product solution, and then optimizes the parameters of the local feature extraction component and the local slope stability identification component. This process not only improves the performance of the model in the slope stability identification task, but also ensures that the local geological analysis platform can provide more accurate and reliable slope stability assessment results.

[0253] In one embodiment, a hardware architecture of a land and resources cloud platform is provided. The land and resources cloud platform may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The land and resources cloud platform includes a first processor, a first memory, a first input / output interface (Input / Output, referred to as I / O) and a first communication interface. Among them, the first processor, the first memory and the first input / output interface are connected through a first system bus, and the first communication interface is connected to the first system bus through the first input / output interface. Among them, the first processor of the land and resources cloud platform is used to provide computing and control capabilities. The first memory of the land and resources cloud platform includes a first non-volatile storage medium and a first internal memory. The first non-volatile storage medium stores a first operating system, a first computer program and a first database. The first internal memory provides an environment for the operation of the first operating system and the first computer program in the first non-volatile storage medium. The database of the land and resources cloud platform is used to store geological monitoring data. The first input / output interface of the land and resources cloud platform is used to exchange information between the first processor and an external device. The first communication interface of the land and resources cloud platform is used to communicate with an external terminal through a network connection. When the first computer program is executed by the first processor, a method for intelligent identification of slope stability is implemented.

[0254] In one embodiment, a hardware architecture of a local geological analysis platform is provided. The local geological analysis platform may be a server, and its internal structure diagram may be as follows: Figure 5As shown. The local geological analysis platform includes a second processor, a second memory, a second input / output interface (Input / Output, referred to as I / O) and a second communication interface. The second processor, the second memory and the second input / output interface are connected through a second system bus, and the second communication interface is connected to the second system bus through the second input / output interface. The second processor of the local geological analysis platform is used to provide computing and control capabilities. The second memory of the local geological analysis platform includes a second non-volatile storage medium and a second internal memory. The second non-volatile storage medium stores a second operating system, a second computer program and a second database. The second internal memory provides an environment for the operation of the second operating system and the second computer program in the second non-volatile storage medium. The database of the local geological analysis platform is used to store geological monitoring data. The second input / output interface of the local geological analysis platform is used to exchange information between the second processor and an external device. The second communication interface of the local geological analysis platform is used to communicate with an external terminal through a network connection. When the second computer program is executed by the second processor, a method for intelligent identification of slope stability is implemented.

[0255] Those skilled in the art will understand that Figure 4 and Figure 5 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the platform to which the solution of the present application is applied. Specifically, it may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

Claims

1. A method for intelligent identification of slope stability, characterized in that: Applied to the land and resources cloud platform, the method includes: The land and resources cloud platform obtains x first characterization arrays, and performs multi-dimensional characterization array extraction on the x first characterization arrays based on the feature extraction component in the cornerstone neural network to obtain second characterization arrays corresponding to each of the x first characterization arrays, where x≥1; the cornerstone neural network also includes a linear mapping component and a slope stability identification component respectively deployed on x local geological analysis platforms; the x first characterization arrays are obtained by extracting features from the slope geological data of the monitored target area through the x local geological analysis platforms respectively based on the deployed local feature extraction components; the local feature extraction components deployed by each local geological analysis platform are obtained based on the optimization of internal configuration parameters of the linear mapping component; Sending one of the x second characterization arrays to the x local geological analysis platforms respectively; wherein the second characterization array acquired by the target local geological analysis platform corresponds to the first characterization array obtained by feature extraction of the target local geological analysis platform; the target local geological analysis platform is any one of the x local geological analysis platforms; If the local feature extraction components and local slope stability identification components respectively deployed by the x local geological analysis platforms meet the local network calibration stop requirements, obtain the target calibration errors respectively sent by the x local geological analysis platforms, wherein the target calibration errors sent by the target local geological analysis platform are constructed by the local slope stability identification components deployed by the target local geological analysis platform and the acquired second characterization array, and the local slope stability identification components deployed by each local geological analysis platform are obtained after optimizing the internal configuration parameters of the slope stability identification components; Based on x target calibration errors, the internal configuration parameters of the feature extraction component are optimized to obtain a collaborative feature extraction component; based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms, a collaborative slope stability identification component corresponding to each local geological analysis platform is generated; at the same time, the collaborative slope stability identification components arranged on the land and resources cloud platform and the target local geological analysis platform are configured to perform slope stability analysis on the slope geological data in the target local geological analysis platform.

2. The method according to claim 1, characterized in that Also includes: Obtaining data element attributes corresponding to each of the x local geological analysis platforms, obtaining multi-dimensional shared data having commonality with the data element attributes, adjusting and optimizing the preset neural network according to the multi-dimensional shared data, and obtaining a cornerstone neural network that reaches a stable state for each dimension; The linear mapping component and the slope stability identification component of the cornerstone neural network are sent to the x local geological analysis platforms, so that each local geological analysis platform optimizes the internal configuration parameters of the deployed linear mapping component based on the slope geological data of the monitored target area and the feature extraction component deployed on the land and resources cloud platform to obtain the local feature extraction component, and optimizes the internal configuration parameters of the deployed slope stability identification component to obtain the local slope stability identification component.

3. The method according to claim 1, characterized in that The land resources cloud platform obtains x first representation arrays, including: Acquire a platform selection ratio, select x local geological analysis platforms from y local geological analysis platforms according to the platform selection ratio, send a calibration command to the x local geological analysis platforms, so that each local geological analysis platform performs linear mapping on the slope geological data based on the local feature extraction component, and obtain a first representation array, wherein 1≤x≤y; Obtaining first representation arrays respectively sent by the x local geological analysis platforms; The feature extraction component based on the cornerstone neural network performs multi-dimensional representation array extraction on the x first representation arrays to obtain second representation arrays corresponding to each of the x first representation arrays, including: Performing multi-channel saliency focusing on the x first representation arrays to obtain saliency focusing arrays corresponding to each of the x first representation arrays; Fully connect the x saliency focus arrays to obtain linear mapping arrays corresponding to the x saliency focus arrays, and generate second representation arrays corresponding to the x first representation arrays according to the x linear mapping arrays.

4. The method according to claim 1, characterized in that The step of optimizing the internal configuration parameters of the feature extraction component based on the x target calibration errors to obtain a collaborative feature extraction component includes: Based on x target calibration errors and internal configuration parameters in the feature extraction component, generating unit calibration errors corresponding to the x target calibration errors respectively, generating a cloud platform bias vector according to the x unit calibration errors and the calibration tuning parameters, optimizing the internal configuration parameters of the feature extraction component according to the cloud platform bias vector, and obtaining a collaborative feature extraction component; The method of generating a collaborative slope stability identification component corresponding to each local geological analysis platform based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms includes: Performing a collaborative integration operation on the internal configuration parameters of the local feature extraction components and the internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms to obtain network adjustment parameters, and sending the network adjustment parameters to the x local geological analysis platforms so that each local geological analysis platform optimizes the parameters of the deployed local feature extraction components and local slope stability identification components according to the network adjustment parameters to obtain collaborative feature extraction components and collaborative slope stability identification components; If the collaborative feature refining component meets the cloud platform network adjustment stopping requirements, the collaborative feature refining component is combined with the collaborative feature extraction component and the collaborative slope stability identification component deployed by each local geological analysis platform to form a collaborative slope stability identification component corresponding to each local geological analysis platform; wherein, the collaborative feature refining component is combined with the collaborative feature extraction component and the collaborative slope stability identification component deployed by the target local geological analysis platform to form a collaborative slope stability identification component corresponding to the target local geological analysis platform.

5. The method according to claim 4, characterized in that The network adjustment parameters include feature extraction iteration parameters and slope stability identification iteration parameters; the internal configuration parameters of the local feature extraction components and the internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms are collaboratively integrated to obtain the network adjustment parameters, including: Obtaining internal configuration parameters of local feature extraction components in the x local geological analysis platforms, performing collaborative integration operations on the internal configuration parameters of the x local feature extraction components, and obtaining feature extraction iteration parameters corresponding to each of the x local geological analysis platforms; the feature extraction iteration parameters are used to send to the corresponding local geological analysis platform, and are configured to perform parameter optimization on the local feature extraction components deployed on the corresponding local geological analysis platform to obtain collaborative feature extraction components; The internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms are obtained, and the internal configuration parameters of the x local slope stability identification components are collaboratively integrated to obtain the slope stability identification iteration parameters corresponding to each of the x local geological analysis platforms; the slope stability identification iteration parameters are used to be sent to the corresponding local geological analysis platform, and are configured to perform parameter optimization on the local slope stability identification components deployed on the corresponding local geological analysis platform to obtain a collaborative slope stability identification component.

6. A method for intelligent identification of slope stability, characterized in that: Applied to a target local geological analysis platform, the method comprises: The target local geological analysis platform performs linear mapping on the slope geological data according to the linear mapping component in the cornerstone neural network to obtain a first characterization array, and sends the first characterization array to the land and resources cloud platform, so that the land and resources cloud platform performs multi-dimensional characterization array extraction on the first characterization array based on the feature extraction component in the cornerstone neural network to obtain a second characterization array; the linear mapping component in the cornerstone neural network is deployed on x local geological analysis platforms, the cornerstone neural network also includes a slope stability identification component for deployment on the x local geological analysis platforms, and the x local geological analysis platforms include the target local geological analysis platform; Acquire the second characterization array sent by the land and resources cloud platform, perform slope stability identification on the second characterization array based on the slope stability identification component to obtain a reasoning identification result, obtain an iterative adjustment error based on the priori mark of the slope geological data and the reasoning identification result, optimize the internal configuration parameters of the linear mapping component and the slope stability identification component based on the iterative adjustment error, and obtain a local feature extraction component and a local slope stability identification component; If both the local feature extraction component and the local slope stability identification component meet the local network calibration stop requirements, the target calibration error generated based on the local slope stability identification component will be sent to the land and resources cloud platform, so that the land and resources cloud platform optimizes the internal configuration parameters of the feature extraction component according to the target calibration errors corresponding to each of the x local geological analysis platforms, and obtains a collaborative feature extraction component. Based on the collaborative feature extraction component and the local feature extraction components and local slope stability identification components in the x local geological analysis platforms, a collaborative slope stability identification component corresponding to each local geological analysis platform is generated; at the same time, the collaborative slope stability identification component arranged on the land and resources cloud platform and the target local geological analysis platform is configured to perform slope stability analysis on the slope geological data in the target local geological analysis platform.

7. The method according to claim 6, characterized in that The target local geological analysis platform linearly maps the slope geological data according to the linear mapping component in the cornerstone neural network to obtain a first representation array, including: Decomposing the slope geological data into i geological data items, performing linear mapping on the i geological data items respectively, and obtaining data item representation arrays corresponding to the i geological data items, i≥1; Embed distribution information of each geological data item in the slope geological data to obtain i distribution representation arrays, fuse the i data item representation arrays and the i distribution representation arrays to obtain a transition representation array, fuse the transition representation array and the identification code array to obtain a first representation array, wherein the identification code array covers the overall feature information of the i data item representation arrays; The step of performing slope stability identification on the second characterization array based on the slope stability identification component to obtain a reasoning identification result includes: Solving the inner product of the second representation array and the score array to obtain a weight representation array, wherein the score array represents a mapping relationship between the second representation array and the priori label set; The weight characterization array and the offset array are fused to obtain a projection array, the projection array is standardized, the reasoning confidence corresponding to each slope stability prior marker in the prior marker set is obtained, and the slope stability prior marker corresponding to the maximum reasoning confidence in the prior marker set is used as the reasoning recognition result of the second characterization array, wherein the offset array is an adaptively adjusted parameter array.

8. The method according to claim 6, characterized in that The method of optimizing the internal configuration parameters of the linear mapping component and the slope stability identification component based on the iterative calibration error to obtain a local feature extraction component and a local slope stability identification component includes: Generate a feature extraction error parameter according to the iterative calibration error and the internal configuration parameter of the linear mapping component, solve the inner product of the feature extraction error parameter and the calibration tuning parameter to obtain a feature extraction iterative bias vector, optimize the internal configuration parameter of the linear mapping component according to the feature extraction iterative bias vector to obtain a local feature extraction component; According to the iterative calibration error and the internal configuration parameters of the slope stability identification component, an identification error parameter is generated, the inner product of the identification error parameter and the calibration tuning parameter is solved to obtain an identification iterative deviation vector, and the internal configuration parameters of the slope stability identification component are optimized according to the identification iterative deviation vector to obtain a local slope stability identification component.

9. The method according to claim 6, characterized in that Also includes: Obtaining network adjustment parameters sent by the land and resources cloud platform; The network adjustment parameters include feature extraction iteration parameters and slope stability identification iteration parameters, wherein the feature extraction iteration parameters are obtained by the land and resources cloud platform performing a collaborative integration operation on the internal configuration parameters of the local feature extraction components in the x local geological analysis platforms, and the slope stability identification iteration parameters are obtained by the land and resources cloud platform performing a collaborative integration operation on the internal configuration parameters of the local slope stability identification components in the x local geological analysis platforms; Solving the inner product of the feature extraction iteration parameter and the first collaborative training iteration weight to obtain a collaborative feature extraction bias vector, optimizing the parameters of the local feature extraction component deployed on the target local geological analysis platform according to the collaborative feature extraction bias vector, and obtaining the collaborative feature extraction component corresponding to the target local geological analysis platform; The inner product of the feature extraction iterative parameter and the second collaborative training iterative weight is solved to obtain a collaborative identification bias vector. Based on the collaborative feature extraction bias vector, the parameters of the local slope stability identification component deployed on the target local geological analysis platform are optimized to obtain the collaborative slope stability identification component corresponding to the target local geological analysis platform.

10. An intelligent identification system for slope stability, characterized in that: The invention comprises a land and resources cloud platform and at least one local geological analysis platform communicatively connected to the land and resources cloud platform, wherein the land and resources cloud platform comprises a first processor; and a first memory, wherein a first computer program is stored in the first memory, and when the first computer program is run by the first processor, the first processor executes the method as described in any one of claims 1 to 5; the local geological analysis platform comprises a second processor; and a second memory, wherein a second computer program is stored in the second memory, and when the second computer program is run by the second processor, the second processor executes the method as described in any one of claims 6 to 9.