An ecological slope support online intelligent detection method and system

Abstract

The application relates to an ecological slope support online intelligent detection method and system, which comprises the following steps: collecting ecological parameters and structural parameters of the slope; performing causal inspection on the two types of parameters to distinguish a causal index set and an independent index; adopting different processing strategies, constructing a feature vector for the causal index set, and obtaining a first result by using a first risk prediction model after deep learning model fusion; performing correlation analysis on the independent index, classifying the independent index, and obtaining a second result by using a second risk prediction model; and combining the two results to obtain a total risk prediction result of the target slope. The system is used for executing the above method. The application predicts the stability risk of the slope by combining the causal relationship between the ecological parameters and the structural parameters of the ecological slope and the correlation between the indexes, so that the causal relationship and the correlation between the indexes can be fully considered during risk prediction, the accuracy of the stability risk prediction is improved, and the accuracy and effectiveness of the online intelligent detection of the ecological slope are improved.

Description

Technical Field

[0001] This application relates to the technical field of slope protection detection, specifically to an online intelligent detection method and system for ecological slope protection. Background Technology

[0002] Traditional slope protection techniques include gravity retaining walls, buttress retaining walls, and cantilever supports, which reinforce slopes by installing retaining walls, anchor cables, and anchor bolts. Ecological slope protection, on the other hand, combines traditional methods with ecological planting techniques. Planting vegetation on the slope surface beautifies the slope and also helps reinforce it and reduce soil erosion. Compared to traditional methods alone, ecological slope protection offers superior protection.

[0003] Inspection of slope support is an important part of slope operation and maintenance. By monitoring parameters such as anchor stress, retaining plate bending moment, shear force, and support structure displacement in real time, the real-time status of the slope support structure can be effectively detected, thereby assessing slope stability.

[0004] Traditional slope protection testing mostly focuses on monitoring the physical and mechanical properties of the support structure. However, for ecological slopes, the traditional support structure is only one part of the ecological slope system. Ecological parameters of the ecological slope system, such as vegetation cover, soil moisture, and soil density, are equally important. More importantly, the causal relationship between ecological parameters and the physical and mechanical properties of the support structure is of great significance for real-time detection and risk prediction of the ecological slope system. Existing slope protection testing systems have not taken this characteristic into account, resulting in low accuracy in the detection of ecological slope protection, which needs to be improved. Summary of the Invention

[0005] To address the problems existing in the prior art, this application aims to provide an online intelligent detection method and system for ecological slope support. This application combines the causal relationship between ecological and structural parameters of ecological slopes and the correlation between indicators to predict slope stability risk. This allows for full consideration of the causal relationship and correlation between indicators during risk prediction, which helps improve the accuracy of stability risk prediction and thus enhances the accuracy and effectiveness of online intelligent detection of ecological slopes.

[0006] The online intelligent detection method for ecological slope protection described in this application includes the following steps:

[0007] S1. Collect ecological and structural parameters of the target slope. The ecological parameters include several ecological indicators related to the ecological state of the target slope, and the structural parameters include several structural indicators related to the stability of the support structure of the target slope.

[0008] S2. Perform a causal test on the ecological parameters and the structural parameters, and mark the ecological indicators and structural indicators that have a causal relationship as a causal indicator set, the causal indicator set also including the lag order; mark the ecological indicators and structural indicators that do not have a causal relationship as independent indicators.

[0009] S3. The causal index set and the independent index are processed using the first processing strategy and the second processing strategy, respectively.

[0010] The first processing strategy includes:

[0011] Based on the causal index set, construct the feature vector corresponding to the causal index set;

[0012] The feature vector is input into a pre-configured deep learning model to perform multi-source data fusion and obtain fused feature data.

[0013] The fused feature data is input into a pre-configured first risk prediction model to obtain a first risk prediction result for the target slope;

[0014] The second processing strategy includes:

[0015] A correlation analysis is performed on each of the independent indicators, and the independent indicators with correlation are grouped into one category to obtain several sets of independent indicators of different types. The sets of independent indicators are then input into a pre-configured second risk prediction model to obtain the second risk prediction result for the target slope.

[0016] S4. Combining the first risk prediction result and the second risk prediction result, obtain the total risk prediction result for the target slope.

[0017] Preferably, in step S1, the ecological indicators include vegetation coverage, average vegetation height, normalized difference vegetation index, soil moisture, soil density, surface runoff direction and surface runoff velocity, and the structural indicators include anchor bolt / anchor cable stress, retaining plate shear force, support structure displacement, slope protection net damage rate and lattice beam crack size.

[0018] Preferably, step S2 specifically includes:

[0019] Data preprocessing is performed on the ecological parameters and the structural parameters;

[0020] The range of the lag order is determined to be [ord] min ,ord max Let the lag order of the structural indicator Y be p, and the lag order of the ecological indicator X be q, where p and q both fall within the range of orders. min ,ordmax ];

[0021] For each pair of ecological indicators X and structural indicators Y, construct both an unrestricted model and a restricted model:

[0022] The unrestricted model is represented as follows:

[0023]

[0024] The constraint model is expressed as follows:

[0025]

[0026] Where t represents time, Y t Y represents the structural index value at the current moment. t-i a represents the lag value of the structural index Y over the past i time steps. i X represents the corresponding coefficient. t-j β represents the lagged value of ecological indicator X over the past j time steps. i Represents the corresponding coefficient, ∈ i Represents the random error term;

[0027] Regression analysis was performed on the unrestricted model and the restricted model respectively to obtain the residual sum of squares (RSS) of the unrestricted model. U And the residual sum of squares (RSS) of the constraint model. R ;

[0028] Calculate the F-statistic using the following formula:

[0029]

[0030] Where n represents the number of samples, i.e. the length of the time series data;

[0031] At a given significance level, the corresponding critical value is obtained by looking up the F-statistic distribution table. If the calculated F-statistic is greater than the critical value, then the ecological indicator X is determined to be a Granger cause of the ecological indicator Y, and the corresponding lag order is recorded. X-Y The causal index set is constructed as (X, Y, ord) X-Y If the calculated F-statistic is not greater than the critical value, then it is determined that there is no causal relationship between the ecological indicator X and the ecological indicator Y. If a certain ecological indicator or structural indicator has no causal relationship with any other indicator, then the ecological indicator or structural indicator is marked as an independent indicator.

[0032] Preferably, in step S3, the training process of the deep learning model includes the following steps:

[0033] Construct sequentially cascaded CNN and LSTM modules;

[0034] Collect multiple sets of time-series data on ecological and structural indicators of slopes;

[0035] For the same group of ecological and structural indicators, Granger causality analysis was used to conduct causal analysis, and the ecological and structural indicators with causal relationships were selected and denoted as the sample causal dataset.

[0036] The sample causal dataset is divided into a training set, a validation set, and a test set. Based on the sample causal dataset, a feature vector corresponding to the sample causal dataset is constructed.

[0037] The feature vectors corresponding to the training set are input into the CNN module, and convolution and pooling operations are performed sequentially to extract feature data;

[0038] The obtained feature data is input into the LSTM module, and the extracted feature data is fused by the LSTM module to output fused features containing causal relationships and lag orders between indicators.

[0039] The parameters of the deep learning model are updated using a loss function and a backpropagation algorithm.

[0040] Repeat the training steps until the model converges or meets the required number of iterations, thus obtaining the deep learning model.

[0041] Preferably, in step S3, the training process of the first risk prediction model includes the following steps:

[0042] A multi-layer fully connected neural network containing an input layer, a hidden layer, and an output layer is constructed to label the corresponding slope risk level for each pair of ecological indicators and structural indicators in the sample causal dataset, and denoted as the true risk level.

[0043] The fused features output by the LSTM module are input into the input layer of the multilayer fully connected neural network;

[0044] The fused features received by the input layer are then passed to multiple hidden layers in sequence. In the hidden layers, nonlinear transformations are performed through activation functions and activation values ​​are calculated. The fused features are then integrated and transformed through multiple hidden layers, and the activation values ​​of the multiple hidden layers are calculated in sequence.

[0045] The activation values ​​of the obtained multi-layer hidden layers are passed to the output layer. The output layer combines the activation values ​​of the multi-layer hidden layers to output the predicted slope risk level, which is denoted as the predicted risk level.

[0046] Based on the difference between the actual risk level and the predicted risk level, the parameters of the multilayer fully connected neural network are updated using a loss function and a backpropagation algorithm.

[0047] Repeat the training steps until the network converges or meets the required number of iterations, thus obtaining the multilayer fully connected neural network.

[0048] Preferably, in step S3, the Pearson correlation coefficient method, K-Means clustering algorithm or principal component analysis method are used to perform correlation analysis and classify each of the independent indicators, and the second risk prediction model is a decision tree model;

[0049] Step S4 includes the following steps:

[0050] The first risk prediction result and the second risk prediction result are numerically normalized, and the numerical normalization result is summed using a weighted summation method. The result obtained is the total risk prediction result.

[0051] Preferably, the method further includes: setting the initial acquisition frequency of both the ecological parameters and the structural parameters to a first frequency; and determining whether the target slope has a stability risk based on the total risk prediction result.

[0052] In response to the stability risk of the target slope, the causal relationships and lag orders (ORD) of the ecological and structural indicators in the causal indicator set are obtained. X-Y The collection frequency of ecological indicators belonging to the aforementioned causal indicator set is increased to a second frequency, and the lag order of ecological indicators and structural indicators is combined. X-Y After the first time interval Time, the sampling frequency of the corresponding structural indicators is increased to a third frequency, wherein the third frequency is greater than the second frequency, and the first time interval Time ∈ [75% * ord]. X-Y ,ord X-Y ].

[0053] This application discloses an online intelligent detection system for ecological slope protection, comprising:

[0054] The acquisition module is used to acquire ecological and structural parameters of the target slope. The ecological parameters include several ecological indicators related to the ecological state of the target slope, and the structural parameters include several structural indicators related to the stability of the support structure of the target slope.

[0055] The causality test module is used to perform causality tests on the ecological parameters and the structural parameters, and to mark the ecological indicators and structural indicators that have a causal relationship as a causal indicator set, which also includes the lag order; and to mark the ecological indicators and structural indicators that do not have a causal relationship as independent indicators.

[0056] The processing module is used to process the causal index set and the independent index using a first processing strategy and a second processing strategy, respectively.

[0057] The first processing strategy includes:

[0058] Based on the causal index set, construct the feature vector corresponding to the causal index set;

[0059] The feature vector is input into a pre-configured deep learning model to perform multi-source data fusion and obtain fused feature data.

[0060] The fused feature data is input into a pre-configured first risk prediction model to obtain a first risk prediction result for the target slope;

[0061] The second processing strategy includes:

[0062] A correlation analysis is performed on each of the independent indicators, and the independent indicators with correlation are grouped into one category to obtain several sets of independent indicators of different types. The sets of independent indicators are then input into a pre-configured second risk prediction model to obtain the second risk prediction result for the target slope.

[0063] The prediction module is used to combine the first risk prediction result and the second risk prediction result to obtain the total risk prediction result for the target slope.

[0064] A computer device according to this application includes a processor and a memory connected by a signal. The memory stores at least one instruction or at least one program. When the at least one instruction or the at least one program is loaded by the processor, it executes the online intelligent detection method for ecological slope support as described above.

[0065] This application discloses a computer-readable storage medium storing at least one instruction or at least one program, which, when loaded by a processor, executes the online intelligent detection method for ecological slope support as described above.

[0066] The online intelligent detection method and system for ecological slope support described in this application has the advantage that it performs causal analysis on the ecological and structural parameters of the target slope, using the causal relationship between these parameters as input information for a first risk prediction model. Furthermore, it performs correlation analysis on the relationships between independent indicators, using the results as input information for a second risk prediction model. This yields both the first and second risk prediction results, which are then combined to obtain the overall risk prediction result. This allows for a comprehensive consideration of the causal relationship and lag between ecological and structural parameters, as well as the correlation between independent indicators, when predicting slope stability risk. This makes the risk prediction for ecological slopes more comprehensive and better suited to the actual conditions of ecological slopes, improving the accuracy of ecological slope risk prediction and thus enhancing the accuracy and effectiveness of online intelligent detection of ecological slopes. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the steps of an online intelligent detection method for ecological slope protection as described in this application;

[0068] Figure 2 This is a schematic diagram of the structure of the computer device described in this embodiment.

[0069] Figure labeling: 101 - Processor, 102 - Memory. Detailed Implementation

[0070] like Figure 1 As shown, the online intelligent detection method for ecological slope protection described in this application includes the following steps:

[0071] S1. Collect ecological and structural parameters of the target slope. The ecological parameters include several ecological indicators related to the ecological state of the target slope, and the structural parameters include several structural indicators related to the stability of the support structure of the target slope.

[0072] Furthermore, the ecological indicators include vegetation coverage, average vegetation height, normalized difference vegetation index, soil moisture, soil density, surface runoff direction, and surface runoff velocity; the structural indicators include anchor bolt / anchor cable stress, retaining plate shear force, support structure displacement, slope protection net damage rate, and lattice beam crack size.

[0073] Specifically, vegetation coverage is calculated by dividing the vegetation coverage area by the total slope area. The vegetation coverage area can be obtained by acquiring images of the slope (e.g., aerial images taken by drones). The images of the vegetation-covered areas are significantly different from those of the bare slope areas, so the area of ​​the vegetation-covered areas can be obtained through image analysis algorithms.

[0074] The average vegetation height is mainly obtained through manual measurement. During slope maintenance, maintenance personnel will conduct regular inspections of the slope. During the inspection, the growth height of the plants can be manually measured. By measuring the plant height in different areas of the slope and calculating the average value, the average vegetation height during that period can be obtained.

[0075] Normalized Difference Vegetation Index (NDVI) is acquired via UAV remote sensing. After acquiring multiple UAV image data, it undergoes processing such as stitching and orthorectification to ensure the spatial accuracy of the images. Then, the red and near-infrared band data in the images are identified and extracted, and then substituted into the NDVI calculation to generate the NDVI result reflecting the vegetation status of the slope.

[0076] Soil moisture can be collected by a moisture sensor inserted into the soil of the slope.

[0077] Soil density mainly relies on manual collection during inspections, such as by using the ring cutter method or wax sealing method to calculate the soil density. Since the soil density changes at a low frequency, a second collection and measurement can be performed after a relatively long interval following a single collection.

[0078] The direction of surface runoff can be obtained through manual observation, while the velocity of surface runoff can be obtained through a velocity sensor placed in the fluid.

[0079] Anchor bolt / cable stress and retaining plate shear force can be obtained by placing force sensors at the monitoring location, such as shear stress sensors commonly used in engineering monitoring.

[0080] Displacement of the support structure can be monitored by installing displacement sensors at the monitoring locations of the support structure.

[0081] The damage rate of the slope protection net can be obtained using aerial images of the aforementioned slope. By using image recognition algorithms to identify the damaged areas in the slope protection net, the area of ​​the damaged area is extracted and divided by the total area of ​​the slope to calculate the damage rate of the slope protection net.

[0082] The crack size of the lattice beam can be calculated by manually taking images of the lattice beam's surface and using existing image crack calculation methods.

[0083] In summary, the various ecological and structural indicators required for ecological slope monitoring can be obtained.

[0084] S2. Perform a causal test on the ecological parameters and the structural parameters, and mark the ecological indicators and structural indicators that have a causal relationship as a causal indicator set, which also includes the lag order; mark the ecological indicators and structural indicators that do not have a causal relationship as independent indicators.

[0085] Step S2 specifically includes:

[0086] The ecological parameters and structural parameters are preprocessed. Specifically, all the aforementioned ecological and structural parameters are time series data with a certain time length, ensuring that all parameters are within the same acquisition period. For parameters with low change frequency, such as soil density, the parameter is considered constant within the acquisition period.

[0087] Since different ecological and structural indicators may have different dimensions and numerical ranges, it is necessary to standardize the data of each indicator in order to eliminate the influence of dimensional differences on the causality test results. In this embodiment, the maximum-minimum normalization method is used to normalize each parameter so that the values ​​of all indicators fall within the range of [0,1], which facilitates subsequent calculations.

[0088] The range of the lag order is determined to be [ord] min ,ord max For example, [1 week, 52 weeks], let p be the lag order of the structural indicator Y and q be the lag order of the ecological indicator X, where p and q are both within the lag order range. min ,ord max ];

[0089] For each pair of ecological indicators X and structural indicators Y, construct both an unrestricted model and a restricted model:

[0090] The unrestricted model is represented as follows:

[0091]

[0092] The restricted model (assuming X is not a Granger cause of Y) is expressed as:

[0093]

[0094] Where t represents time, Y t Y represents the structural index value at the current moment. t-i a represents the lag value of the structural index Y over the past i time steps. i X represents the corresponding coefficient. t-j β represents the lagged value of ecological indicator X over the past j time steps. i Represents the corresponding coefficient, ∈ i Represents the random error term;

[0095] Regression analysis was performed on the unrestricted model and the restricted model respectively to obtain the residual sum of squares (RSS) of the unrestricted model. U And the residual sum of squares (RSS) of the constraint model. R ;

[0096] Calculate the F-statistic using the following formula:

[0097]

[0098] Where n represents the number of samples, i.e. the length of the time series data;

[0099] At a given significance level, such as the commonly used 5% significance level, the corresponding critical value is obtained by looking up the F-statistic distribution table. If the calculated F-statistic is greater than the critical value, the null hypothesis is rejected (assuming X is not a Granger cause of Y), and the ecological indicator X is determined to be a Granger cause of the ecological indicator Y. The corresponding lag order q is recorded, and this lag order q is the lag order between the ecological indicator X and the corresponding structural indicator Y. X-Y This indicates that when the ecological indicator X changes significantly, it will be displayed in ord X-Y The celestial influence affects the structural index Y. The causal index set is constructed as (X, Y, ord) X-Y ).

[0100] For example, causal analysis was conducted on soil moisture, an ecological indicator, and anchor stress, a structural indicator, as mentioned above.

[0101] Real-time soil moisture and anchor stress are collected. For example, the soil moisture value within a certain monitoring period is 25% obtained through a moisture sensor, assuming the minimum soil moisture value is 10% and the maximum value is 40%. Normalization is performed using the maximum-minimum normalization method.

[0102]

[0103] Similarly, the anchor stress is normalized using the maximum-minimum normalization method.

[0104] Let soil moisture be X and anchor stress be Y. Construct the unrestricted and restricted models described above. Preset lag order values ​​p = 3 and q = 2. Calculate the sum of squared residuals of the two models as described above and use the F-statistic to determine the critical value, thus judging whether a Granger causal relationship exists between soil moisture X and anchor stress Y. Specifically, using information criterion methods, such as the Akaike Information Criterion (AIC) or the Schwarz Information Criterion (SIC), calculate the information criterion values ​​for different lag orders. Based on the obtained information criterion values, determine the optimal lag order. Generally, the lag order with the smallest information criterion value is selected as the optimal lag order; for example, given the lag order values ​​p = 3 and q = 2 above.

[0105] The above steps determine that soil moisture is the Granger cause of anchor stress, with a lag order of 2 weeks. In this embodiment, only the lag effect of ecological indicators on structural indicators is analyzed, so p=3 is discarded.

[0106] If the calculated F-statistic is not greater than the critical value, it is determined that there is no causal relationship between the ecological indicator X and the ecological indicator Y. If a certain ecological indicator or structural indicator has no causal relationship with any other indicator, it is marked as an independent indicator.

[0107] S3. The causal index set and the independent index are processed using the first processing strategy and the second processing strategy, respectively.

[0108] The first processing strategy includes:

[0109] Based on the causal index set, construct the feature vector corresponding to the causal index set;

[0110] The feature vector is input into a pre-configured deep learning model to perform multi-source data fusion and obtain fused feature data.

[0111] Furthermore, in step S3, the training process of the deep learning model includes the following steps:

[0112] 1. Construct sequentially cascaded CNN and LSTM modules.

[0113] CNN Module: Convolutional Neural Networks (CNNs) are primarily used to extract local features from data. When building a CNN module, it is necessary to determine the number and parameters of convolutional layers and pooling layers.

[0114] Convolutional layers: Convolutional layers contain multiple convolutional kernels, each performing a sliding convolution operation on the input data to extract different features. The size, number, and stride of the convolutional kernels are important parameters that need to be set. For example, the kernel size can be set to 3x3 or 5x5, and the number of kernels can be adjusted according to the complexity of the data and the needs of feature extraction.

[0115] Pooling layers: Pooling layers are used to reduce the dimensionality of feature maps, decrease computational cost, and enhance the robustness of the model. Common pooling operations include max pooling and average pooling. The size of the pooling window and the stride are also parameters that need to be set.

[0116] LSTM Module: Long Short-Term Memory (LSTM) networks are a special type of recurrent neural network capable of handling long-term dependencies in sequential data. When constructing an LSTM module, the number of LSTM units, i.e., the number of neurons in the hidden layers, needs to be determined. LSTM units control the flow of information through gating mechanisms (input gate, forget gate, and output gate), thereby effectively capturing temporal dependencies in sequential data.

[0117] Cascaded approach: The output of the CNN module is used as the input of the LSTM module, realizing the sequential cascading of the two modules. The CNN module first extracts local features from the input data, and then the LSTM module performs time-series processing and fusion on the extracted features.

[0118] 2. Collect multiple sets of time series data on ecological and structural indicators of slopes.

[0119] Ecological indicators include vegetation cover, average vegetation height, normalized difference vegetation index, soil moisture, soil density, surface runoff direction, and surface runoff velocity.

[0120] Structural parameters include anchor bolt / anchor cable stress, retaining plate shear force, support structure displacement, slope protection net damage rate, and lattice beam crack size.

[0121] Time series data: This involves collecting continuous observations of these indicators over a period of time to form time series data. The data interval can be set according to actual needs and the accuracy of the monitoring equipment, such as collecting data daily, weekly, or monthly.

[0122] 3. For the same group of ecological and structural indicators, Granger causality analysis is used to conduct causal analysis, and the ecological and structural indicators with causal relationships are selected and recorded as the sample causal dataset.

[0123] The methods for Granger causality analysis and determination of lag order can be understood by referring to the above description, and will not be repeated here.

[0124] Sample causal dataset: Ecological and structural indicators with causal relationships are selected, and their corresponding time series data are combined into a sample causal dataset.

[0125] 4. Divide the sample causal dataset into a training set, a validation set, and a test set, and construct the feature vector corresponding to the sample causal dataset based on the sample causal dataset.

[0126] Dataset partitioning: The causal dataset is divided into training, validation, and test sets according to a certain ratio. Common partition ratios are 70%-15%-15% or 80%-10%-10%. The training set is used for learning the model's parameters, the validation set is used to adjust the model's hyperparameters (such as learning rate, batch size, etc.), and the test set is used to evaluate the model's final performance.

[0127] Feature vector construction: Construct corresponding feature vectors based on the ecological and structural indicators in the sample causal dataset.

[0128] 5. Input the feature vectors corresponding to the training set into the CNN module, and perform convolution and pooling operations in sequence to extract feature data.

[0129] Convolution operation: The feature vectors corresponding to the training set are input into the convolutional layers of the CNN module. Each convolutional kernel performs a sliding convolution operation on the feature vectors to generate the corresponding feature map. Convolution operation can extract local features from the feature vectors, and different convolutional kernels can extract different types of features.

[0130] Pooling operation: The feature map output from the convolutional layer undergoes a pooling operation to reduce its dimensionality. For example, max pooling selects the maximum value in each pooling window as the output, thereby reducing the size of the feature map while retaining important feature information.

[0131] Feature data extraction: After multiple convolution and pooling operations, the CNN module outputs the extracted feature data. This feature data contains local feature information from the sample causal dataset, providing input for the subsequent LSTM module.

[0132] 6. Input the obtained feature data into the LSTM module, and fuse the extracted feature data through the LSTM module to output fused features containing causal relationships and lag orders between indicators.

[0133] LSTM Unit Processing: The feature data output from the CNN module is input into the LSTM module. The LSTM module processes the input feature data through a gating mechanism. The forget gate determines how much information from the previous cell state needs to be forgotten, the input gate determines how much information from the current input needs to be added to the cell state, and the output gate determines how much information from the current cell state needs to be output.

[0134] Feature fusion: During sequence data processing, the LSTM module fuses feature data from different time steps to capture temporal and causal relationships between indicators. By continuously updating cell and hidden states, the LSTM module can learn long-term dependency information in the sequence data.

[0135] Fusion Feature Output: The final output of the LSTM module contains fused features that include causal relationships between indicators and lag orders. These fused features combine the local features extracted by the CNN module and the temporal dependencies captured by the LSTM module, thus providing a more comprehensive reflection of the feature information of the causal dataset.

[0136] 7. Update the parameters of the deep learning model using the loss function and backpropagation algorithm.

[0137] Loss Function: Choosing an appropriate loss function measures the difference between the model's predictions and the true labels. For regression problems, commonly used loss functions include mean squared error (MSE) and mean absolute error (MAE); for classification problems, commonly used loss functions include cross-entropy loss. The appropriate loss function should be selected based on the specific task requirements.

[0138] Backpropagation Algorithm: The backpropagation algorithm is a method used to calculate gradients and update model parameters. It calculates the model's loss value based on the loss function, and then uses the backpropagation algorithm to calculate the gradient of the loss function with respect to the model parameters. Based on the calculated gradient, optimization algorithms (such as stochastic gradient descent, Adam, etc.) are used to update the model parameters, causing the value of the loss function to gradually decrease.

[0139] 8. Repeat the training steps until the model converges or the required number of iterations is met, thus obtaining the deep learning model.

[0140] Iterative training: Repeat steps 5-7, continuously inputting the feature vectors from the training set into the model for training and updating the model's parameters. Each iteration gradually improves the model's performance and gradually reduces the value of the loss function.

[0141] Convergence criterion: During training, the convergence of the model is determined by observing the value of the loss function and the performance metrics of the validation set (such as accuracy, mean squared error, etc.). If the value of the loss function no longer decreases significantly, or the performance metrics of the validation set no longer improve, it indicates that the model has converged.

[0142] Iteration Limit: In addition to determining the end of training based on convergence, a maximum number of iterations can be set. When the maximum number of iterations is reached, training stops even if the model has not fully converged. The resulting trained deep learning model can then utilize the causal relationships and lag order information of ecological and structural indicators to perform slope-related predictions and analyses.

[0143] The deep learning model can be constructed through the above steps. When performing actual slope detection, by inputting the feature vector corresponding to the causal index set after Granger analysis, the deep learning model can capture the causal relationship in the data and generate fused feature information containing the causal relationship between the indicators and the lag order.

[0144] The fused feature data is input into a pre-configured first risk prediction model to obtain a first risk prediction result for the target slope;

[0145] The first risk prediction model is constructed according to the following steps:

[0146] 1. Construct a multi-layer fully connected neural network and label it with the actual risk level.

[0147] Construct a multi-layer fully connected neural network.

[0148] A multilayer fully connected neural network consists of an input layer, hidden layers, and an output layer.

[0149] Input layer: The number of neurons in this layer depends on the dimensionality of the input data. In this embodiment, the input data is the fused feature output by the LSTM module, so the number of neurons in the input layer is consistent with the dimensionality of the fused feature. For example, if the fused feature is a 3-dimensional vector, then the number of neurons in the input layer is 3.

[0150] Hidden layers: Multiple hidden layers can be set. The number of hidden layers and the number of neurons in each hidden layer are hyperparameters that need to be adjusted. Generally, increasing the number of hidden layers and neurons can improve the model's expressive power, but it also increases the risk of overfitting. It is generally advisable to try a smaller number of hidden layers (e.g., 2-3 layers) and a moderate number of neurons (e.g., 30-100 neurons per layer) first, and then adjust them based on the training results.

[0151] Output layer: The number of neurons in the output layer depends on the number of categories in the prediction task. Since this is a prediction of slope risk level, assuming that the slope risk level is divided into three levels: high, medium, and low, the number of neurons in the output layer is 3.

[0152] For each pair of ecological and structural indicators in the sample causal dataset, the corresponding slope risk level is labeled based on professional knowledge, historical data, or expert experience, and these labeled risk levels are recorded as the true risk levels. For example, when the soil moisture is too high and the anchor stress increases abnormally, based on experience, it is judged that the slope is at a high risk level, and the sample corresponding to this pair of indicators is labeled as "high risk".

[0153] 2. Input fused features into the input layer.

[0154] The fused features output from the LSTM module are input into the input layer of a multi-layer fully connected neural network. Each fused feature vector corresponds to one sample, and each neuron in the input layer receives the value of the corresponding dimension from the fused feature vector.

[0155] 3. The transfer and transformation of fusion features in the hidden layer.

[0156] The calculation process of the hidden layer.

[0157] The fused features received at the input layer are sequentially passed to multiple hidden layers. In each hidden layer, the neurons perform the following calculations:

[0158] Weighted summation: Each neuron receives the outputs of all neurons in the previous layer, multiplies these outputs by their corresponding weights, then sums the products and adds a bias term. The mathematical expression is:

[0159]

[0160] Where, z j It is the weighted sum of the current neurons, w ij It is the weight from the i-th neuron in the previous layer to the current neuron, x i It is the output of the i-th neuron in the previous layer, b j It is the bias term of the current neuron, and n is the number of neurons in the previous layer.

[0161] Nonlinear transformation: weighted sum z j The activation value 'a' of the current neuron is obtained by performing a non-linear transformation through an activation function. j Common activation functions include ReLU (Rectified Linear Unit), Sigmoid, and Tanh.

[0162] Multiple hidden layers progressively integrate and transform the fused features. Each hidden layer computes based on the output of the previous layer, extracting higher-level feature representations. For example, the first hidden layer might extract some basic feature combinations, while subsequent hidden layers further combine and abstract these basic features to form more representative features. In this way, multiple hidden layers can deeply mine and transform fused features.

[0163] 4. The output layer outputs the predicted risk level.

[0164] The activation values ​​from multiple hidden layers are passed to the output layer. Each neuron in the output layer performs a weighted sum of the input activation values, and then transforms the output into a probability distribution using a suitable activation function (such as the Softmax function). Assuming the output layer has three neurons corresponding to high, medium, and low risk levels, the Softmax function converts the output of each neuron into a probability value, and the sum of these three probability values ​​is 1. The risk level corresponding to the neuron with the highest probability value is the predicted slope risk level, which is denoted as the predicted risk level.

[0165] 5. Update network parameters based on differences.

[0166] Calculate the loss function:

[0167] To measure the difference between the true risk level and the predicted risk level, an appropriate loss function needs to be selected. For multi-class classification problems, the commonly used loss function is the cross-entropy loss function.

[0168] Backpropagation algorithm updates parameters:

[0169] Based on the calculated loss function value, the backpropagation algorithm is used to calculate the gradient of the loss function with respect to all parameters (including weights and biases) in the neural network. The backpropagation algorithm calculates the gradient layer by layer, starting from the output layer, using the chain rule.

[0170] 6. Repeat the training until convergence or the required number of iterations is met.

[0171] Repeat steps 2-5, continuously inputting the fused features into the network for training and updating the network parameters. During training, it is important to monitor the value of the loss function and the model's performance on the validation set.

[0172] Convergence criterion: If the value of the loss function no longer decreases significantly after multiple iterations, or if the model's performance on the validation set (such as accuracy, F1 score, etc.) no longer improves, it indicates that the model has converged.

[0173] Iteration limit: A maximum number of iterations can also be set. When the training reaches the maximum number of iterations, training stops even if the model has not fully converged. The final result is a trained multi-layer fully connected neural network, which can be used to predict the risk level of slopes.

[0174] The above steps can be used to construct a first risk prediction model. When actually conducting slope detection, the fusion feature information extracted by the aforementioned deep learning model can be input into the first risk prediction model to predict the risk level of the target slope.

[0175] The second processing strategy includes:

[0176] A correlation analysis is performed on each of the independent indicators, and the independent indicators with correlation are grouped into one category to obtain several sets of independent indicators of different types. The sets of independent indicators are then input into a pre-configured second risk prediction model to obtain the second risk prediction result for the target slope. The Pearson correlation coefficient method, K-Means clustering algorithm, or principal component analysis method are used to perform correlation analysis and classify each of the independent indicators. The second risk prediction model is a decision tree model.

[0177] Specifically, the decision tree model selects information gain as the splitting criterion. For example, at the root node, the information gain of each indicator is calculated. Assume the current dataset has 80 sample points, and slope risk is divided into three levels: high, medium, and low, with 20, 30, and 30 points respectively. Taking average vegetation height as an example, it is used as the splitting feature. Based on different average vegetation height thresholds, the dataset is divided into subsets, and the information entropy and information gain before and after the split are calculated. Assume that after the split, two subsets are obtained: subset 1 has 30 sample points with risk levels of 5, 15, and 10; subset 2 has 50 sample points with risk levels of 15, 15, and 20.

[0178] Calculate the information entropy before partitioning and the conditional entropy after partitioning, and calculate the information gain. Compare the information gain of each indicator, select the indicator with the largest information gain as the partitioning feature of the root node, and then recursively construct the subtree.

[0179] The process continues to split each subset until a stopping condition is met, such as a subset having fewer than 5 samples or all samples belonging to the same category. To prevent overfitting, a post-pruning method is used; after the decision tree is built, it is pruned based on the performance on the test set.

[0180] The independent indicator set data from the training set is input into the decision tree model for training. The model learns the relationship between different combinations of independent indicators and slope risk. For example, when the average vegetation height is high, the soil density is high, the retaining wall shear force is moderate, and the slope protection net damage rate is low, the model may determine that the slope is at a low risk level (similarly, each set of sample data is pre-labeled based on expert experience).

[0181] The trained decision tree model is evaluated using a test set, and metrics such as accuracy, recall, and F1 score are calculated. Based on the evaluation results, the parameters of the decision tree model are adjusted until the evaluation results meet the requirements, thus obtaining the second risk prediction model.

[0182] S4. Combining the first risk prediction result and the second risk prediction result, obtain the total risk prediction result for the target slope.

[0183] The first risk prediction result and the second risk prediction result are numerically normalized, and the numerical normalization result is summed using a weighted summation method. The result obtained is the total risk prediction result.

[0184] By combining the prediction results of the two models through a weighted summation, the first risk prediction result focuses on the causal relationship between ecological and structural indicators, while the second risk prediction result focuses on the correlation between independent indicators. This allows the total risk prediction result to integrate the correlation between various indicators, making the prediction result more accurate and comprehensive.

[0185] Furthermore, the method also includes: setting the initial acquisition frequency of both the ecological parameters and the structural parameters to a first frequency, i.e., the initial acquisition frequency; and determining whether the target slope has a stability risk based on the total risk prediction result, for example by comparing the total risk prediction result with a preset risk threshold, and if it is greater than the risk threshold, determining that the target slope has a stability risk.

[0186] In response to the stability risk of the target slope, i.e., the prediction that the target slope may suffer a disaster in the future, the causal relationships and lag orders (ORD) of the ecological and structural indicators in the causal indicator set are obtained. X-Y The collection frequency of ecological indicators belonging to the aforementioned causal indicator set is increased to a second frequency, and the lag order of ecological indicators and structural indicators is combined. X-Y After the first time interval Time, the sampling frequency of the corresponding structural indicators is increased to a third frequency, wherein the third frequency is greater than the second frequency, and the first time interval Time ∈ [75% * ord]. X-Y ,ord X-Y In this step, when a potential disaster risk is predicted for the target slope, the frequency of monitoring data collection needs to be increased to ensure timely detection of slope anomalies. Since the ecological and structural indicators within the causal index set have a causal relationship (e.g., soil moisture and anchor stress), soil moisture induces changes in anchor stress. Therefore, the collection frequency of upstream soil moisture is increased first, for example, by increasing the collection frequency of the soil moisture sensor. Furthermore, by utilizing the hysteresis order between soil moisture and anchor stress, the timing for increasing the anchor stress collection frequency can be set. For example, based on hysteresis order analysis, changes in soil moisture induce significant changes in anchor stress after two weeks. Therefore, the collection frequency of anchor stress is increased approximately two weeks after the current time. Downstream anchor stress more directly reflects the stability of the target slope, so its collection frequency is higher than that of soil moisture. Through this design, the collection frequency of each indicator can be reasonably adjusted based on the causal relationship and hysteresis order. This allows for timely adjustment of the collection frequency based on predicted risks to effectively monitor the target slope while rationally controlling the collection frequency and reducing unnecessary energy consumption of the detection system.

[0187] This application conducts causal analysis on the ecological and structural parameters of the target slope, using the causal relationship between these parameters as input information for a first risk prediction model. By performing correlation analysis on independent indicators, the results are used as input information for a second risk prediction model. The first and second risk prediction results are obtained separately, and combined to obtain the overall risk prediction result. This approach comprehensively considers the causal relationship and lag between ecological and structural parameters, as well as the correlation between independent indicators, when predicting slope stability risk. This makes the risk prediction for ecological slopes more comprehensive and better suited to the actual conditions of ecological slopes, improving the accuracy of ecological slope risk prediction and thus enhancing the accuracy and effectiveness of online intelligent detection of ecological slopes.

[0188] This application conducts causal analysis on the ecological and structural parameters of the target slope, using the causal relationship between these parameters as input information for a first risk prediction model. By performing correlation analysis on independent indicators, the results are used as input information for a second risk prediction model. The first and second risk prediction results are obtained separately, and combined to obtain the overall risk prediction result. This approach comprehensively considers the causal relationship and lag between ecological and structural parameters, as well as the correlation between independent indicators, when predicting slope stability risk. This makes the risk prediction for ecological slopes more comprehensive and better suited to the actual conditions of ecological slopes, improving the accuracy of ecological slope risk prediction and thus enhancing the accuracy and effectiveness of online intelligent detection of ecological slopes.

[0189] This application also provides an online intelligent detection system for ecological slope protection, including:

[0190] The acquisition module is used to acquire ecological and structural parameters of the target slope. The ecological parameters include several ecological indicators related to the ecological state of the target slope, and the structural parameters include several structural indicators related to the stability of the support structure of the target slope.

[0191] The causality test module is used to perform causality tests on the ecological parameters and the structural parameters, and to mark the ecological indicators and structural indicators that have a causal relationship as a causal indicator set, which also includes the lag order; and to mark the ecological indicators and structural indicators that do not have a causal relationship as independent indicators.

[0192] The processing module is used to process the causal index set and the independent index using a first processing strategy and a second processing strategy, respectively.

[0193] The first processing strategy includes:

[0194] Based on the causal index set, construct the feature vector corresponding to the causal index set;

[0195] The feature vector is input into a pre-configured deep learning model to perform multi-source data fusion and obtain fused feature data.

[0196] The fused feature data is input into a pre-configured first risk prediction model to obtain a first risk prediction result for the target slope;

[0197] The second processing strategy includes:

[0198] A correlation analysis is performed on each of the independent indicators, and the independent indicators with correlation are grouped into one category to obtain several sets of independent indicators of different types. The sets of independent indicators are then input into a pre-configured second risk prediction model to obtain the second risk prediction result for the target slope.

[0199] The prediction module is used to combine the first risk prediction result and the second risk prediction result to obtain the total risk prediction result for the target slope.

[0200] The online intelligent detection system for ecological slope protection proposed in this application belongs to the same inventive concept as the method described above, and can be understood with reference to the above description, which will not be repeated here.

[0201] like Figure 2 As shown, this embodiment also provides a computer device, including a processor 101 and a memory 102 connected via a bus signal. The memory 102 stores at least one instruction or at least one program segment. When the at least one instruction or the at least one program segment is loaded by the processor 101, it executes the online intelligent detection method for ecological slope support as described above. The memory 102 can be used to store software programs and modules. The processor 101 executes various functional applications by running the software programs and modules stored in the memory 102. The memory 102 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for functions, etc.; the data storage area may store data created according to the use of the device, etc. In addition, the memory 102 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 102 may also include a memory controller to provide the processor 101 with access to the memory 102.

[0202] The methods and embodiments provided in this application can be executed in a computer terminal, server, or similar computing device; that is, the aforementioned computer device may include a computer terminal, server, or similar computing device. The internal structure of the computer device may include, but is not limited to, a processor, a network interface, and memory. The processor, network interface, and memory within the computer device can be connected via a bus or other means.

[0203] The processor 101 (or CPU, Central Processing Unit) is the computing and control core of the computer device. The network interface may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.). The memory 102 is the storage device in the computer device used to store programs and data. It is understood that the memory 102 here can be a high-speed RAM storage device, or a non-volatile storage device, such as at least one disk storage device; optionally, it can also be at least one storage device located remotely from the processor 101. The memory 102 provides storage space that stores the operating system of the electronic device, which may include, but is not limited to: Windows (an operating system), Linux (an operating system), Android (a mobile operating system), iOS (a mobile operating system), etc., and this application does not limit this; furthermore, the storage space also stores one or more instructions suitable for being loaded and executed by the processor 101, which may be one or more computer programs (including program code). In the embodiments of this specification, the processor 101 loads and executes one or more instructions stored in the memory 102 to implement the online intelligent detection method for ecological slope support described in the above method embodiments.

[0204] This application also provides a computer-readable storage medium storing at least one instruction or at least one program segment. When the at least one instruction or at least one program segment is loaded by the processor 101, it executes the online intelligent detection method for ecological slope support as described above. The aforementioned computer-readable storage medium carries one or more programs, and when the one or more programs are executed, the method according to the embodiments of this application is implemented.

[0205] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0206] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this application.

Claims

1. An ecological slope support online intelligent detection method, characterized in that, Includes the following steps: S1. Collect ecological and structural parameters of the target slope. The ecological parameters include several ecological indicators related to the ecological state of the target slope, and the structural parameters include several structural indicators related to the stability of the support structure of the target slope. S2. Perform a causal test on the ecological parameters and the structural parameters, and mark the ecological indicators and structural indicators that have a causal relationship as a causal indicator set, the causal indicator set also including the lag order; mark the ecological indicators and structural indicators that do not have a causal relationship as independent indicators. S3. The causal index set and the independent index are processed using the first processing strategy and the second processing strategy, respectively. The first processing strategy includes: Based on the causal index set, construct the feature vector corresponding to the causal index set; The feature vector is input into a pre-configured deep learning model to perform multi-source data fusion and obtain fused feature data. The fused feature data is input into a pre-configured first risk prediction model, which is a multi-layer fully connected neural network including an input layer, a hidden layer and an output layer, to obtain a first risk prediction result for the target slope. The second processing strategy includes: A correlation analysis is performed on each of the independent indicators, and the independent indicators with correlation are grouped into one category to obtain several sets of independent indicators of different types. The sets of independent indicators are then input into a pre-configured second risk prediction model to obtain a second risk prediction result for the target slope. The second risk prediction model is a decision tree model. S4. Combining the first risk prediction result and the second risk prediction result, obtain the total risk prediction result for the target slope. Perform numerical normalization on the first risk prediction result and the second risk prediction result, and perform weighted summation on the numerical normalization result. The calculated result is the total risk prediction result.

2. The online intelligent detection method for ecological slope support according to claim 1, characterized in that, In step S1, the ecological indicators include vegetation coverage, average vegetation height, normalized difference vegetation index, soil moisture, soil density, surface runoff direction and surface runoff velocity, and the structural indicators include anchor bolt / anchor cable stress, retaining plate shear force, support structure displacement, slope protection net damage rate and lattice beam crack size.

3. The online intelligent detection method for the ecological slope support according to claim 1 or 2, characterized in that, Step S2 specifically includes: Data preprocessing is performed on the ecological parameters and the structural parameters; The order range of the order of the lag order is , the structure index is set , the lag order of the self is , the ecological index is , the lag order of the self is , , , and all belong to the order range ; For each pair of ecological indicators and structural indicators , unrestricted and restricted models are constructed respectively: The unrestricted model is represented as follows: ； The constraint model is expressed as follows: ； wherein, denotes time, denotes a structure indicator value at the current time instant, denotes a structure indicator in the past lagged values of the last time steps, denotes the corresponding coefficient, denotes an ecological indicator in the past lagged values of the last time steps, denotes the corresponding coefficient, denotes a random error term; Regression analyses were performed on the unrestricted model and the restricted model respectively to obtain the sum of squared residuals of the unrestricted model. and the sum of squared residuals of the constrained model. ; Calculate using the following formula Statistic: ； in, This indicates the number of samples, i.e., the length of the time series data. At a given significance level, by searching Obtain the corresponding critical value from the statistical distribution table. If the calculated value is... If the statistic is greater than the threshold value, then the ecological indicator is judged to be... For this ecological indicator Find the Granger cause and record the corresponding lag order. The causal index set is constructed as follows: If the calculated result If the statistic is not greater than the threshold value, then the ecological indicator is judged to be... With this ecological indicator If there is no causal relationship, then an ecological indicator or structural indicator is marked as an independent indicator.

4. The online intelligent detection method for ecological slope protection according to claim 3, characterized in that, In step S3, the training process of the deep learning model includes the following steps: Construct sequentially cascaded CNN and LSTM modules; Collect multiple sets of time-series data on ecological and structural indicators of slopes; For the same group of ecological and structural indicators, Granger causality analysis was used to conduct causal analysis, and the ecological and structural indicators with causal relationships were selected and denoted as the sample causal dataset. The sample causal dataset is divided into a training set, a validation set, and a test set. Based on the sample causal dataset, a feature vector corresponding to the sample causal dataset is constructed. The feature vectors corresponding to the training set are input into the CNN module, and convolution and pooling operations are performed sequentially to extract feature data; The obtained feature data is input into the LSTM module, and the extracted feature data is fused by the LSTM module to output fused features containing causal relationships and lag orders between indicators. The parameters of the deep learning model are updated using a loss function and a backpropagation algorithm. Repeat the training steps until the model converges or meets the required number of iterations, thus obtaining the deep learning model.

5. The online intelligent detection method for ecological slope protection according to claim 4, characterized in that, In step S3, the training process of the first risk prediction model includes the following steps: A multi-layer fully connected neural network containing an input layer, a hidden layer, and an output layer is constructed to label the corresponding slope risk level for each pair of ecological indicators and structural indicators in the sample causal dataset, and denoted as the true risk level. The fused features output by the LSTM module are input into the input layer of the multilayer fully connected neural network; The fused features received by the input layer are then passed to multiple hidden layers in sequence. In the hidden layers, nonlinear transformations are performed through activation functions and activation values ​​are calculated. The fused features are then integrated and transformed through multiple hidden layers, and the activation values ​​of the multiple hidden layers are calculated in sequence. The activation values ​​of the obtained multi-layer hidden layers are passed to the output layer. The output layer combines the activation values ​​of the multi-layer hidden layers to output the predicted slope risk level, which is denoted as the predicted risk level. Based on the difference between the actual risk level and the predicted risk level, the parameters of the multilayer fully connected neural network are updated using a loss function and a backpropagation algorithm. Repeat the training steps until the network converges or meets the required number of iterations, thus obtaining the multilayer fully connected neural network.

6. The online intelligent detection method for ecological slope protection according to claim 5, characterized in that, In step S3, the correlation analysis and classification of each independent index are performed using the Pearson correlation coefficient method, K-Means clustering algorithm, or principal component analysis.

7. The online intelligent detection method for ecological slope protection according to claim 6, characterized in that, Also includes: The initial acquisition frequency for both the ecological parameters and the structural parameters is preset to be a first frequency; Based on the overall risk prediction results, determine whether the target slope has any stability risk; In response to the stability risk of the target slope, the causal relationships and lag orders of the ecological and structural indicators in the causal indicator set are obtained. The collection frequency of ecological indicators belonging to the aforementioned causal indicator set is increased to a second frequency, and the lag order of ecological indicators and structural indicators is combined. In the first time interval The sampling frequency of the corresponding structural indicators is then increased to a third frequency, wherein the third frequency is greater than the second frequency, and the first time interval is... .

8. An online intelligent detection system for ecological slope protection, characterized in that, include: The acquisition module is used to acquire ecological and structural parameters of the target slope. The ecological parameters include several ecological indicators related to the ecological state of the target slope, and the structural parameters include several structural indicators related to the stability of the support structure of the target slope. The causality test module is used to perform causality tests on the ecological parameters and the structural parameters, and to mark the ecological indicators and structural indicators that have a causal relationship as a causal indicator set, which also includes the lag order; and to mark the ecological indicators and structural indicators that do not have a causal relationship as independent indicators. The processing module is used to process the causal index set and the independent index using a first processing strategy and a second processing strategy, respectively. The first processing strategy includes: Based on the causal index set, construct the feature vector corresponding to the causal index set; The feature vector is input into a pre-configured deep learning model to perform multi-source data fusion and obtain fused feature data. The fused feature data is input into a pre-configured first risk prediction model, which is a multi-layer fully connected neural network including an input layer, a hidden layer and an output layer, to obtain a first risk prediction result for the target slope. The second processing strategy includes: A correlation analysis is performed on each of the independent indicators, and the independent indicators with correlation are grouped into one category to obtain several sets of independent indicators of different types. The sets of independent indicators are then input into a pre-configured second risk prediction model to obtain a second risk prediction result for the target slope. The second risk prediction model is a decision tree model. The prediction module combines the first risk prediction result and the second risk prediction result to obtain the total risk prediction result for the target slope. It performs numerical normalization on the first risk prediction result and the second risk prediction result, and then performs a weighted summation on the numerical normalization result. The result obtained is the total risk prediction result.

9. A computer device comprising a processor and a memory connected by signals, characterized in that, The memory stores at least one instruction or at least one program segment, which is executed by the processor when loaded, according to any one of claims 1-7, the online intelligent detection method for ecological slope support.

10. A computer-readable storage medium having stored thereon at least one instruction or at least one program, characterized in that, When the at least one instruction or the at least one program segment is loaded by the processor, the online intelligent detection method for ecological slope support as described in any one of claims 1-7 is executed.

Images