Earth and rockfill dam workability evaluation method and system based on multi-source heterogeneous data input

Abstract

The invention discloses an earth and rockfill dam workability evaluation method and system based on multi-source heterogeneous data input, and relates to the technical field of hydropower engineering structures and artificial intelligence, and the method comprises the steps: carrying out the extraction of image and text data key features of an existing earth and rockfill dam, and obtaining a comprehensive multi-source heterogeneous data set; based on a One-hot encoder, performing cross-channel fusion processing on the comprehensive multi-source heterogeneous data set to obtain an input data set and an output label data set of the earth and rockfill dam image-text heterogeneous high-dimensional feature data; according to an input data set and an output label data set of the earth and rockfill dam image-text heterogeneous high-dimensional feature data, a TTAO-EDconv-LSTM-Attention prediction model is trained; and inputting the image and text data of the target earth and rockfill dam into the trained prediction model to obtain a prediction result of the target earth and rockfill dam. According to the invention, quantitative coupling evaluation can be carried out on deformation stability, dam slope stability, seepage stability and anti-seismic safety of the earth and rockfill dam.

Description

Technical Field

[0001] The present application relates to the field of hydropower engineering structure and artificial intelligence technology, and in particular to a method and system for evaluating the working performance of an earth-rock dam based on multi-source heterogeneous data input. Background Art

[0002] As an important water conservancy and hydropower engineering facility, the safe operation of earth-rock dams is crucial for flood control, irrigation, water supply and power generation. Due to the influence of the natural environment and factors such as material aging, earth-rock dams may have various safety hazards, such as deformation, seepage, landslides, etc. If these problems are not discovered and handled in time, they may lead to serious safety accidents. Therefore, long-term and continuous prediction and evaluation of the properties of earth-rock dams is the key to ensure their safe operation.

[0003] However, most of the existing technologies are based on finite element calculations to analyze the properties of various dimensions such as deformation or seepage. On the one hand, finite element calculations involve the establishment of finite element models and the conduct of complex static and dynamic simulation analysis. The calculation process is time-consuming and labor-intensive, and it is difficult to quickly obtain the evaluation results of the dam's working properties; on the other hand, it is difficult to form a comprehensive quantitative evaluation result for the evaluation results of different dimensions, so it is impossible to fully reflect the working properties of the dam. With the rapid development of artificial intelligence technology, the construction of a proxy model for the safety and stability of earth-rock dams based on deep learning algorithms provides a new path for the rapid and accurate evaluation of the working properties of earth-rock dams. Summary of the invention

[0004] The purpose of this application is to provide a method and system for evaluating the working performance of earth-rock dams based on multi-source heterogeneous data input, which can perform quantitative coupling evaluation on the deformation stability, dam slope stability, seepage stability and seismic safety of earth-rock dams.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides a method for evaluating the working performance of an earth-rock dam based on multi-source heterogeneous data input, the method comprising:

[0007] Extract key features of image and text data from existing earth-rock dam design data to obtain a comprehensive multi-source heterogeneous data set; the comprehensive multi-source heterogeneous data set includes a dam deformation stability data set, a dam slope stability data set, a dam seepage data set, and a dam seismic safety data set; the comprehensive multi-source heterogeneous data set is used to train a TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is used to predict the working performance of earth-rock dams;

[0008] Based on the One-hot encoder, the comprehensive multi-source heterogeneous dataset is subjected to cross-channel fusion processing to obtain an input dataset and an output label dataset of the earth-rock dam image-text heterogeneous high-dimensional feature data;

[0009] According to the input data set and output label data set of heterogeneous high-dimensional feature data of earth-rock dam images and texts, the TTAO-EDconv-LSTM-Attention prediction model is trained to obtain a trained TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is a deep learning model that introduces TTAO mechanism, EDconv and Attention mechanism on the basis of the original LSTM neural network;

[0010] The image and text data of the target earth-rock dam are input into the trained TTAO-EDconv-LSTM-Attention prediction model to obtain the prediction results of the target earth-rock dam; the prediction results include deformation stability, dam slope stability, seepage stability and seismic safety.

[0011] Optionally, after obtaining the prediction result of the target earth-rock dam, the method further includes:

[0012] The prediction results are transformed into probability distribution based on the Softmax multi-index coupling evaluator to obtain the dam safety assessment result; the dam safety assessment result is composed of the accuracy, precision, recall rate and F1 score calculated by the confusion matrix.

[0013] Optionally, the prediction result is converted into a probability distribution based on a Softmax multi-index coupling evaluator, specifically including:

[0014] According to the formula Performing probability distribution conversion on the prediction result;

[0015] Among them, y i ∈y 1 ,y 2 ,y 3 ,y 4 ,y i is the predicted value of the i-th dimension, y 1 is the predicted value of deformation stability, y 2 is the predicted value of dam slope stability, y 3 is the predicted value of seepage stability, y 4 is the predicted value of earthquake safety, P(y i ) is the predicted probability corresponding to the i-th dimension, and j represents the predicted dimension.

[0016] Optionally, the calculation formula for the accuracy of confusion matrix calculation in the dam safety assessment result is:

[0017] The calculation formula for the accuracy of the dam safety assessment results is:

[0018] The calculation formula of the recall rate in the dam safety assessment result is:

[0019] The calculation formula of F1-Score in the dam safety assessment result is:

[0020] In the formula, TP is a true positive example, FP is a false positive example, FN is a false negative example, TN is a true negative example, Recall is the recall rate, Precision is the accuracy rate, and F1-Score is the harmonic average of precision and recall rate.

[0021] Optionally, the key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including:

[0022] Construct a dam deformation stability dataset, including:

[0023] Collect multi-source data of earth-rock dams and extract geometric parameters of the earth-rock dams; the geometric parameters include dam body shape, size, slope, soil type, permeability coefficient, density, cohesion, internal friction angle, and water level change;

[0024] According to the extracted geometric parameters, the finite element model of the earth-rock dam is established based on the finite element calculation software;

[0025] Performing deformation analysis on the finite element model of the earth-rock dam to obtain the maximum tensile stress, cracking conditions and settlement of the earth-rock dam during completion under different conditions;

[0026] A deformation stability dataset is constructed based on the maximum tensile stress, cracking and settlement of the earth-rock dam under different conditions.

[0027] Optionally, the key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including:

[0028] Construct a dam slope stability dataset, including:

[0029] Based on the limit equilibrium method, the shape and position of the sliding surface are defined;

[0030] According to the formula The dam slope stability analysis is performed on the earth-rock dam finite element model to obtain a dam slope stability data set.

[0031] Optionally, the key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including:

[0032] Construct a dam body seepage data set, including:

[0033] Performing seepage analysis on the finite element model of the earth-rock dam to generate key indicators of seepage volume, pressure head and hydraulic gradient;

[0034] A dam seepage data set is constructed based on the generated key indicators of dam profile, water level, permeability coefficient, seepage rate, pressure head and hydraulic gradient.

[0035] Optionally, the key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including:

[0036] Construct a dam seismic safety data set, including:

[0037] Performing dynamic time history analysis on the finite element model of the earth-rock dam to obtain key indicators of dynamic safety factor, liquefaction area and residual deformation;

[0038] Based on the key indicators of dynamic safety factor, liquefaction area and residual deformation, a dam seismic safety dataset is constructed.

[0039] In the second aspect, the present application provides an earth-rock dam working performance evaluation system based on multi-source heterogeneous data input, comprising:

[0040] A comprehensive multi-source heterogeneous data set construction module is used to extract key features of image and text data from existing earth-rock dam design data to obtain a comprehensive multi-source heterogeneous data set; the comprehensive multi-source heterogeneous data set includes a dam deformation stability data set, a dam slope stability data set, a dam seepage data set, and a dam seismic safety data set; the comprehensive multi-source heterogeneous data set is used to train a TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is used to predict the working performance of earth-rock dams;

[0041] A fusion processing module is used to perform cross-channel fusion processing on the comprehensive multi-source heterogeneous data set based on a One-hot encoder to obtain an input data set and an output label data set of earth-rock dam image-text heterogeneous high-dimensional feature data;

[0042] A model training module is used to train a TTAO-EDconv-LSTM-Attention prediction model according to an input data set and an output label data set of heterogeneous high-dimensional feature data of earth-rock dam images and texts, so as to obtain a trained TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is a deep learning model that introduces TTAO mechanism, EDconv and Attention mechanism on the basis of the original LSTM neural network;

[0043] The prediction module is used to input the image and text data of the target earth-rock dam into the trained TTAO-EDconv-LSTM-Attention prediction model to obtain the prediction results of the target earth-rock dam; the prediction results include deformation stability, dam slope stability, seepage stability and seismic safety.

[0044] Optionally, it also includes:

[0045] The conversion module is used to convert the prediction results into probability distribution based on the Softmax multi-index coupling evaluator to obtain the dam safety assessment result; the dam safety assessment result is composed of the accuracy, precision, recall rate and F1 score calculated by the confusion matrix.

[0046] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0047] The present application provides a method and system for evaluating the working performance of earth-rock dams based on multi-source heterogeneous data input, which forms a multi-source heterogeneous data set by extracting the key features of images and texts of earth-rock dam design data. The data set includes LSTM deformation stability, dam slope stability, seepage and seismic safety data sets, which are used to train the TTAO-EDconv-LSTM-Attention prediction model to predict the working performance of earth-rock dams. The multi-source heterogeneous data sets are fused using a One-hot encoder to form input and output label data sets, and then the TTAO-EDconv-LSTM-Attention model is trained. The model is a deep learning model that combines the TTAO mechanism, EDconv and Attention mechanism. The image and text data of the target earth-rock dam are input into the trained model to obtain prediction results, including deformation stability, dam slope stability, seepage stability and seismic safety. The method provided in the present application can quantitatively evaluate the deformation stability, dam slope stability, seepage stability and seismic safety of earth-rock dams. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 A flowchart of a method for evaluating the working performance of an earth-rock dam based on multi-source heterogeneous data input is provided in accordance with an embodiment of the present application.

[0050] Figure 2 An internal structure diagram of a memory unit in an LSTM provided in one embodiment of the present application.

[0051] Figure 3 A schematic diagram of the structure of an earth-rock dam working performance evaluation system based on multi-source heterogeneous data input provided in one embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0054] Embodiment 1

[0055] like Figure 1 As shown, this embodiment provides a method for evaluating the working performance of an earth-rock dam based on multi-source heterogeneous data input, including:

[0056] Step 101: extract key features of image and text data from existing earth-rock dam design data to obtain a comprehensive multi-source heterogeneous data set; the comprehensive multi-source heterogeneous data set includes a dam deformation stability data set, a dam slope stability data set, a dam seepage data set and a dam seismic safety data set; the comprehensive multi-source heterogeneous data set is used to train a TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is used to predict the working performance of earth-rock dams.

[0057] Step 102: Based on the One-hot encoder, cross-channel fusion processing is performed on the comprehensive multi-source heterogeneous dataset to obtain an input dataset and an output label dataset of earth-rock dam image-text heterogeneous high-dimensional feature data.

[0058] Step 103: According to the input data set and the output label data set of the earth-rock dam image-text heterogeneous high-dimensional feature data, the TTAO-EDconv-LSTM-Attention prediction model is trained to obtain a trained TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is a deep learning model that introduces the TTAO mechanism, EDconv and Attention mechanism on the basis of the original LSTM neural network.

[0059] Step 104: Input the image and text data of the target earth-rock dam into the trained TTAO-EDconv-LSTM-Attention prediction model to obtain the prediction result of the target earth-rock dam; the prediction result includes deformation stability, dam slope stability, seepage stability and seismic safety.

[0060] In some embodiments, when executing step 101, the specific steps may be as follows:

[0061] 1) Construct the dam deformation stability data set, including:

[0062] Multi-source data of earth-rock dams are collected, and geometric parameters of the earth-rock dams are extracted; the geometric parameters include dam body shape, size, slope, soil type, permeability coefficient, density, cohesion, internal friction angle, and water level change.

[0063] According to the extracted geometric parameters, the finite element model of the earth-rock dam is established based on the finite element calculation software.

[0064] The deformation analysis of the finite element model of the earth-rock dam is carried out to obtain the maximum tensile stress, cracking conditions and settlement of the earth-rock dam during the completion period under different conditions.

[0065] According to the maximum tensile stress, cracking and settlement of earth-rock dam under different conditions, a dam deformation stability dataset was constructed.

[0066] Specifically, when constructing the dam deformation stability data set, we collect and organize multi-source data related to earth-rock dams, obtain the geometric parameters of earth-rock dams, including dam shape, size, slope, etc.; geological characteristics: soil type, permeability coefficient, density, cohesion, internal friction angle, etc.; hydrological data: water level change, and then use the seepage analysis module in the finite element calculation software to establish the earth-rock dam model, draw the dam outline according to the dam slope shape and material partition in the drawing information, and determine the material properties to complete the modeling work; then clearly specify the conditions at the inner and outer boundaries of the calculation domain, including support conditions, load conditions, etc. Select a suitable solver and convergence standard to ensure the stability of the calculation process and the accuracy of the results; according to the processed earth-rock dam finite element model, deformation analysis is carried out in the finite element environment, and by simulating different working conditions and load combinations, the maximum tensile stress, cracking and settlement of the earth-rock dam under different conditions are calculated. The key indicators of the deformation analysis are sorted into a data set as the deformation stability input data set for training LSTM.

[0067] 2) Construct a dam slope stability dataset, including:

[0068] Based on the finite element model of the earth-rock dam, the shape and position of the sliding surface are defined.

[0069] Based on the limit equilibrium method, according to the formula The dam slope stability analysis is performed on the earth-rock dam finite element model to obtain a dam slope stability data set.

[0070] Specifically, the earth-rock dam data model used for seepage analysis is imported into the finite element dam slope stability module; the shape and position of the sliding surface are defined, usually a circular sliding surface or a non-circular sliding surface is used, and factors such as the geometric shape of the dam body, material properties, and external loads need to be considered; and an appropriate limit equilibrium method (Morgenstern-Price) is selected for dam slope stability analysis.

[0071] 3) Construct the dam body seepage data set, including:

[0072] A seepage analysis is performed on the finite element model of the earth-rock dam to generate key indicators such as date, water level, permeability coefficient, seepage rate, pressure head and hydraulic gradient.

[0073] A dam body seepage data set was constructed based on the generated date, water level, permeability coefficient, seepage rate, pressure head and hydraulic gradient key indicators.

[0074] Specifically, collect and organize multi-source data related to earth-rock dams to obtain the geometric parameters of the earth-rock dams, including dam shape, size, slope, etc.; geological characteristics: soil type, permeability coefficient, density, cohesion, internal friction angle, etc.; hydrological data: water level changes; use the seepage analysis module in the finite element calculation software to establish an earth-rock dam model, draw the dam body outline and determine the material properties according to the dam slope shape and material partition in the drawing information to complete the modeling work; clearly specify the conditions at the inner and outer boundaries of the calculation domain, including head boundaries, flow boundaries, etc., and reasonably plan the finite element grid layout to ensure that the grid density is sufficient to capture the details of the seepage field; use the prepared model to perform seepage analysis in the finite element calculation environment to generate key indicator data sets of date, water level, permeability coefficient, seepage flow, pressure head, and hydraulic gradient as the seepage input and output label data sets for training LSTM.

[0075] 4) Performing dynamic time history analysis on the finite element model of the earth-rock dam to obtain key indicators of dynamic safety factor, liquefaction area and residual deformation.

[0076] Based on the key indicators of dynamic safety factor, liquefaction area and residual deformation, a dam seismic safety dataset is constructed.

[0077] Specifically, a suitable numerical integration method is selected to solve the structural dynamic equations, and the time step and total analysis time are determined to ensure calculation accuracy and efficiency. Then, the seismic load is set according to the design requirements, which can be represented by the acceleration time history curve. The time resolution of the load curve in the figure is ensured to match the time step of the numerical integration. Key indicators such as maximum tensile stress, maximum settlement, settlement during completion, and deformation at key positions are extracted from the dynamic time history analysis results to calculate the dynamic safety factor (FS), which is used to evaluate the stability of the dam slope and identify the liquefaction area. The location where liquefaction may occur is determined by analyzing the changes in pore water pressure and effective stress. The residual deformation is calculated, that is, the permanent deformation of the structure after the dynamic load. After calculation, the dynamic safety factor, liquefaction area, and residual deformation data set are constructed as the seismic safety data set of the dam body.

[0078] Specifically, the dynamic safety factor (FS) is calculated to evaluate the stability of the dam slope. The liquefaction area is identified and the location where liquefaction may occur is determined by analyzing the changes in pore water pressure and effective stress. The residual deformation, that is, the permanent deformation of the structure after the dynamic load, is calculated. After calculation, the dynamic safety factor, liquefaction area, and residual deformation data set are constructed as the seismic safety data set of the dam body. Dynamic time history analysis (DTHA) is a numerical simulation method used to evaluate the response of a structure under earthquake, wind load or other dynamic loads.

[0079] Dynamic time history analysis is based on the fundamental equation of structural dynamics, namely Newton's second law:

[0080] Safety factor (FS): A dimensionless number used to assess the stability of a dam slope, defined as the ratio of the sliding resistance to the sliding driving force, as follows:

[0081] Then, after completing the collection of the data set, data cleaning and standardization are performed to remove outliers, fill in missing values, and identify and process duplicate data to ensure data integrity and consistency, and convert data with different features into a unified scale. The Z-score standardization method is usually used to make the mean of each feature 0 and the standard deviation 1, thereby eliminating the impact of the dimension and facilitating subsequent feature fusion and model training; the calculation formula for Z-score standardization is as follows: Among them, z is the standardized value; x is the original data value; μ is the mean of the data set; σ is the standard deviation of the data set.

[0082] Extract key output features, fuse image data with encoded text data across channels, use data fusion technology to integrate data from different sources into a unified framework, and construct a comprehensive heterogeneous dataset.

[0083] In some embodiments, when executing step 102, the specific steps may be as follows:

[0084] The image data and text data in the dataset, such as water level value, permeability coefficient, etc., are preprocessed. This includes cleaning features with high correlation and using a one-hot encoder for cross-channel fusion processing. In addition, normalization is performed on different input features and output parameters to construct the corresponding earth-rock dam image-text heterogeneous high-dimensional feature data input and output label dataset. Finally, these data are processed into a nonlinear time series data format.

[0085] The specific steps are as follows: First, the image data is resized, normalized, and enhanced; then, the text data is segmented, stop words are removed, and stems are extracted. After the text data is converted into a vector form using a One-hot encoder, the image features are concatenated with the text features to generate a fused feature vector. Then, these fused feature vectors are subjected to practical normalization; the time series specifications of the comprehensive heterogeneous data sets are adjusted to be consistent, synchronized and regularized to 1 hour, and the data sets are merged into 3 hours and converted into a nonlinear time series data format. Finally, the characteristic parameters of the earth-rock dam with different input features and outputs are normalized to construct the corresponding high-dimensional feature input and output label data sets.

[0086] In some embodiments, when executing step 103, the specific steps may be as follows:

[0087] Construct a prediction model of earth-rock dam behavior based on long short-term memory (LSTM) neural network (such as Figure 2 As shown in the figure, the adaptive training mechanism and multi-step prediction capability are combined; the global optimization capability of TTAO and the feature extraction capability of EDconv are introduced to enhance the generalization ability of the model. The Attention mechanism can give different weights to different variables of the model, thereby highlighting the key feature information. Therefore, the TTAO-EDconv-LSTM-Attention optimizer is constructed, and the early stopping method, a regularization technology, is used to avoid the overfitting problem in the deep learning model training process. The specific steps can be as follows:

[0088] S31. Construct a prediction model for earth-rock dam performance based on long short-term memory (LSTM) neural network, combine it with adaptive training mechanism and multi-step prediction capability, and define a loss function to measure the error between the prediction result and the actual result of the model. Including:

[0089] Specifically, the method for constructing a long short-term memory neural network agent model is as follows:

[0090] Initialize the LSTM network parameters. Specifically, use nn.LSTM to create a single-layer LSTM, batch_first = True to ensure that the input tensor format is (batch_size, seq_len, input_size); then, use nn.Linear to initialize a linear classifier; finally, use nn.init.xavier_uniform_ to perform Xavieruniform initialization on the weights and biases of the LSTM layer and classifier.

[0091] S32. Define mean square error as the loss function loss_fn = nn.MSELoss()py.

[0092] Assume that the model prediction result is: The true value is: y t =[y t1 ,y t2 ,y t3 ,y t4 ], where y t1 ,y t2 ,y t3 ,y t4 They represent the deformation stability, slope stability, seepage stability and seismic safety of earth-rock dams respectively. Then the loss function can be defined according to the mean square error as:

[0093]

[0094] Among them, y ti is the true value of the i-th dimension, is the predicted value of the model; N is the number of samples in the data set; is the weight parameter of the model; λ is the regularization coefficient used to prevent overfitting.

[0095] S33. Update network weights using the back-propagation algorithm.

[0096] Back propagation is to calculate the partial derivative (gradient) of the error with respect to the weight layer by layer through the chain rule, and then pass it back from the output layer to the hidden layer and then to the input layer. This process uses the gradient descent method to update the value of each weight to minimize the loss function. According to the gradient descent algorithm, the value of each weight is updated according to a certain learning rate.

[0097]

[0098] Where η is the learning rate.

[0099] S34. Apply TTAO-EDconv-LSTM-Attention to further improve model performance.

[0100] The TTAO (Time-Time Adaptive Optimization) mechanism can more accurately capture the stability changes of earth-rock dams at different times by adaptively adjusting the input weights and time steps of the time series. This mechanism optimizes the processing method of time series data and improves the model's ability to respond to time changes. TTAO assigns different weighting factors to different time steps, allowing the model to make more accurate predictions of important time points (such as emergencies, drastic changes, etc.). The learning rate is automatically adjusted according to the error changes during the training process, thereby avoiding overfitting or gradient disappearance problems and accelerating model convergence.

[0101] EDconv (dilated convolutional network) is a special convolution operation in convolutional neural networks (CNNs). It increases the receptive field by inserting holes (i.e. dilation) in the convolution kernel. It can efficiently capture local patterns and long-range dependencies in time series data and improve the ability of feature extraction.

[0102] The Attention mechanism can automatically assign weights to different variables, helping the model focus on feature variables that have a greater impact on the prediction results. By introducing the Attention mechanism, the model can better respond to changes in key features and improve prediction accuracy.

[0103] The calculation formula for Attention weight is generally:

[0104]

[0105] Among them, Q is the query vector, which represents the current task or the current moment’s demand; K is the key vector, which represents all input features, equivalent to a candidate set; V is the value vector, which represents the feature vector of all inputs; QK T is the dot product of Q and K, indicating the similarity between them; It is a scaling factor to prevent the inner product from being too large and causing the gradient to disappear.

[0106] The TTAO-EDconv-LSTM-Attention optimizer has higher optimization efficiency, enhanced time-dependent modeling capabilities, and provides accurate multi-dimensional predictions.

[0107] Construct the TTAO-EDconv-LSTM-Attention optimizer and use the regularization technique of early stopping to avoid overfitting problems in the deep learning model training process. The details are as follows:

[0108] S35. Avoid overfitting by early stopping.

[0109] Early Stopping is a commonly used regularization technique that aims to avoid overfitting problems during deep learning model training. It prevents the model from overfitting the training set data by monitoring the model's validation set performance during training and stopping training in a timely manner. Specifically, when the model's performance on the validation set begins to decline, training is terminated early to ensure that the model's performance on the training set and validation set is as close as possible, thereby improving its generalization ability. The core idea of ​​the early stopping method is to continuously monitor the model's loss or accuracy on the validation set during training, and decide whether to stop training based on whether the validation set loss increases, whether the validation set loss improves, and whether the model weight with the smallest validation set loss is maintained during training.

[0110] The modeling process of the forget gate, input gate, and output gate inside the LSTM structural unit is as follows:

[0111] S351. The forget gate controls how much osmotic pressure monitoring data can be accumulated from the memory cells at the previous moment to the memory cells at the current moment. The expression is as follows:

[0112] f 1 =σ(W f [h t-1 ,x t ]+b f ).

[0113] S352. The input gate combines the current memory and long-term memory of LSTM to form a new unit state to mine the temporal features in the seepage pressure monitoring data:

[0114] i t=σ(W t ·[h t-1 ,x t ]+b t ).

[0115] The initial end of the current input The state is shown in the following formula, which is calculated based on the last output of the osmotic pressure effect and the current input:

[0116]

[0117] Current time unit C t The state is shown as follows:

[0118]

[0119] The final output of S353.LSTM is determined by the output gate and the cell state:

[0120] σ t =σ(W 0 [h t-1 ,x 1t ]+b 0 ).

[0121] h t =o t tanh(C t ).

[0122] Where: h t-1 and h t are the output of the cell at the previous moment and the output of the current cell respectively; f t 、i t , C t , o t are the values ​​of the forget gate, input gate, initial cell state at the current moment, cell state at the current moment, and output gate respectively; σ is the Sigmoid function; W and b are the corresponding weight coefficients and bias terms respectively.

[0123] Therefore, the structure of the model is: [input data (image features + text features)] → [EDconv layer (feature extraction)] → [LSTM layer (time series modeling)] → [Attention mechanism (weighting of important features)] → [fully connected layer] → [output layer (multi-step prediction)].

[0124] In some embodiments, when executing step 104, the specific steps may be as follows:

[0125] The LSTM (Long Short-Term Memory) neural network can capture long-term dependencies in time series data and make predictions. Specifically, the LSTM network can predict the performance of earth-rock dams in four key dimensions, including deformation stability, dam slope stability, seepage stability, and seismic safety. In order to achieve a comprehensive evaluation of these multi-dimensional data, the Softmax multi-index coupling evaluator is introduced to convert the output of the model into a probability distribution, and then the confusion matrix is ​​used to calculate indicators such as accuracy, precision, recall rate, and F1 score. The quantitative evaluation results are output through coupling, thereby achieving a comprehensive evaluation of the overall stability of the earth-rock dam.

[0126] The specific steps are as follows:

[0127] S41. Use the trained LSTM model to predict new time series data.

[0128] S411 performs data preprocessing by reading data, creating sample sets, analyzing sample sets, and normalizing samples.

[0129] S412 defines the LSTM model for model training and finally performs model prediction.

[0130] In addition, after obtaining the prediction results, it also includes:

[0131] S42. Input the prediction results into the Softmax multi-index coupling evaluator.

[0132] In order to achieve a comprehensive analysis of multi-dimensional prediction results, the Softmax multi-index coupling evaluator is used. This evaluator couples multi-dimensional data based on the Softmax function and comprehensively processes the prediction values ​​of each dimension output by the LSTM model.

[0133] The Softmax function is often used in multi-classification problems by converting different prediction values ​​into a probability distribution. In the prediction of earth-rock dams, the Softmax evaluator standardizes and weights the prediction values ​​of each dimension and outputs a comprehensive dam safety assessment result.

[0134] For the earth-rock dam model built in this embodiment, the overall stability assessment of the earth-rock dam can be divided into the following categories: normal (stable), potential risk (need to be monitored) and unstable (need to be reinforced). The Softmax function maps the predicted output of each dimension of the built model to a probability value between 0 and 1, and the sum of these probability values ​​is 1; therefore, the network will assign a probability to each possible stability state (such as "normal", "unstable", "critical", etc.).

[0135] S43. The final evaluation score is obtained by integrating the probability distribution of each dimension.

[0136] This embodiment has four prediction dimensions: y 1 ,y 2 ,y 3 ,y 4 (deformation stability, dam slope stability, seepage stability, seismic safety), the Softmax function converts it into a probability distribution:

[0137]

[0138] Among them, y i is the predicted value of each dimension, P(y i ) is the predicted probability corresponding to the dimension, and j represents all predicted dimensions; in this way, the Softmax function converts the prediction result of each dimension into a standardized probability value, thereby providing a quantitative comprehensive evaluation of the overall stability of the earth-rock dam.

[0139] The Softmax formula is:

[0140] in, is the exponential function value of the i-th element in vector z, and j is the index of all categories.

[0141] The confusion matrix is ​​then used to calculate metrics such as accuracy, precision, recall, and F1 score.

[0142] Confusion matrix calculation accuracy: The ratio of the number of samples predicted correctly to the total number of samples.

[0143] Accuracy: The proportion of samples predicted to be positive that are actually positive.

[0144] Recall: The proportion of samples predicted to be positive among samples that are actually positive.

[0145] F1 score: The harmonic mean of precision and recall is used to comprehensively evaluate the predictive ability of the model.

[0146] S44. Determine the overall stability of the earth-rock dam based on the evaluation score:

[0147] S441. The comprehensive evaluation score processes multiple safety dimensions of earth-rock dams (such as deformation, dam slope stability, seepage stability and seismic safety) through LSTM fusion; there is a certain correlation between these dimensions. By fusing different LSTM models or information from different feature sources, the accuracy and stability of the prediction can be effectively improved.

[0148] S442. Fusion can be achieved through model-level fusion. By coupling the probability of each dimension output by the Softmax evaluator, a comprehensive stability score can be obtained.

[0149]

[0150] Where S is the comprehensive stability score of the earth-rock dam; P(y i ) is the predicted probability of the i-th dimension; is the weight of each dimension.

[0151] Different weights can be assigned to each dimension according to its impact on the stability of earth-rock dams. Combining the prediction output of the LSTM neural network and the comprehensive analysis results of the Softmax evaluator, a multi-dimensional comprehensive coupling evaluation of earth-rock dams can be achieved; this not only provides individual prediction results for each dimension, but also obtains an overall safety assessment score through coupling analysis, thus providing a scientific basis for the maintenance and decision-making of earth-rock dams.

[0152] S443. Fusion analysis can also be implemented through feature fusion; monitoring data from different sources (such as deformation, dam slope stability, seepage stability and seismic safety) can be sent to multiple LSTM models as multiple input features; each LSTM model learns the time series information of different features, and then the hidden states of these models (for example, the output of the last time step) are spliced ​​together to form a fused feature vector, which is then input into another LSTM network for comprehensive prediction.

[0153]

[0154] in, They are the hidden states of each LSTM model at the last time step. After concatenating these hidden states, a new input feature X is formed. fused , and input into the subsequent LSTM layer for final prediction.

[0155] S444. In the final simulation, if both deformation stability and dam slope stability show a high probability of instability, but seepage stability and seismic safety are within the normal range, the analysis may need to pay special attention to the structural stability of the dam body.

[0156] In addition, after obtaining the prediction result of the target earth-rock dam, it also includes:

[0157] The prediction results are transformed into probability distribution based on the Softmax multi-index coupling evaluator to obtain the dam safety assessment result; the dam safety assessment result is composed of the accuracy, precision, recall rate and F1 score calculated by the confusion matrix.

[0158] Embodiment 2

[0159] like Figure 3 As shown, this embodiment provides an earth-rock dam working performance evaluation system based on multi-source heterogeneous data input, including:

[0160] The comprehensive multi-source heterogeneous data set construction module 301 is used to extract key features of image and text data from existing earth-rock dam design data to obtain a comprehensive multi-source heterogeneous data set; the comprehensive multi-source heterogeneous data set includes a dam body deformation stability data set, a dam slope stability data set, a dam body seepage data set and a dam body seismic safety data set; the comprehensive multi-source heterogeneous data set is used to train a TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is used to predict the working performance of earth-rock dams.

[0161] The fusion processing module 302 is used to perform cross-channel fusion processing on the comprehensive multi-source heterogeneous data set based on a one-hot encoder to obtain an input data set and an output label data set of earth-rock dam image-text heterogeneous high-dimensional feature data.

[0162] The model training module 303 is used to train the TTAO-EDconv-LSTM-Attention prediction model according to the input data set and the output label data set of the heterogeneous high-dimensional feature data of the earth-rock dam image and text, and obtain the trained TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is a deep learning model that introduces the TTAO mechanism, EDconv and Attention mechanism on the basis of the original LSTM neural network.

[0163] The prediction module 304 is used to input the image and text data of the target earth-rock dam into the trained TTAO-EDconv-LSTM-Attention prediction model to obtain the prediction result of the target earth-rock dam; the prediction result includes deformation stability, dam slope stability, seepage stability and seismic safety.

[0164] The system further includes:

[0165] The conversion module is used to convert the prediction results into probability distribution based on the Softmax multi-index coupling evaluator to obtain the dam safety assessment result; the dam safety assessment result is composed of the accuracy, precision, recall rate and F1 score calculated by the confusion matrix.

[0166] In summary, this application has the following technical effects:

[0167] 1) Improve the accuracy of earth-rock dam performance prediction; This application combines the TTAO-EDconv-LSTM-Attention optimizer for model training, which significantly enhances the prediction accuracy of earth-rock dam performance. The synergy of each module enables the model to more deeply understand and capture the temporal characteristics, spatial characteristics and information at key moments in the earth-rock dam monitoring data, thereby effectively improving the accuracy and stability of the prediction. Through the TTAO (time-time adaptive optimization) mechanism and the Attention mechanism, the model can dynamically adjust the strategy according to different input features, improving the adaptability to multi-source heterogeneous data and complex environmental changes. Optimize the training process and improve training efficiency. The dynamic weighting of the convolutional layer of EDconv and the Attention mechanism reduces the dependence of the traditional LSTM model on a large number of time steps, improves the convergence speed of the model, and makes training on multi-source data and large-scale data sets more efficient.

[0168] 2) Provide a more accurate basis for safety assessment of earth-rock dams. With the improvement of the prediction accuracy of the model, this application can provide a more scientific and accurate basis for the safety assessment and risk management of earth-rock dams. Through the long-term prediction of the dam state, the model can timely discover potential safety hazards (such as dam settlement, leakage, water level abnormalities, etc.), provide decision makers with timely early warning information, and help take preventive measures in advance to ensure the safe operation of the dam.

[0169] 3) Flexible application to different dam bodies and environmental conditions. The deep learning model used in this application has strong adaptability and scalability. It can be adjusted according to the characteristics of different dam bodies and environmental conditions and is applicable to different types of earth-rock dams. For different dam bodies, the model can effectively improve the prediction accuracy through adaptive learning and provide targeted comprehensive evaluation.

[0170] 4) This application can provide an intelligent and precise evaluation method for the performance monitoring of earth-rock dams. Compared with the traditional earth-rock dam performance monitoring method, this application combines multi-source heterogeneous data input, LSTM model, and TTAO-EDconv-LSTM-Attention optimizer to effectively improve the accuracy and timeliness of earth-rock dam performance prediction, thereby realizing dynamic real-time monitoring of dam structure and operation status, and providing decision makers with a scientific and objective safety assessment basis.

[0171] 5) Through the intelligent coupling evaluation method of this application, a comprehensive risk assessment of earth-rock dams can be achieved, and early warnings can be issued in a timely manner based on the prediction results. The risk early warning function helps to identify dam body abnormalities (such as settlement, cracks, water level abnormalities, etc.) in advance and take preventive measures to prevent catastrophic accidents and ensure the long-term stable operation of water conservancy and hydropower projects.

[0172] 6) Through the intelligent performance prediction model of this application, the status of the reservoir dam can be analyzed in real time, and accurate dam performance prediction can be generated by combining historical data, sensor data and environmental data. Large-scale earth-rock dam performance group monitoring This application method can uniformly analyze and evaluate multiple dams by constructing a unified multi-source heterogeneous data input and intelligent prediction model, thereby providing support for the comprehensive management of large-scale water conservancy and hydropower systems. This method is particularly suitable for applications in scenarios such as earth-rock dam projects.

[0173] 7) This application provides innovative ideas and tools for interdisciplinary research and application. Through the integration of multidisciplinary technologies, better results can be achieved in earth-rock dam monitoring, risk assessment, intelligent decision-making, etc., and the deep integration of civil engineering, environmental monitoring and artificial intelligence can be promoted.

[0174] 8) This application uses an intelligent prediction system to predict the future state change trend of the earth-rock dam body and provide early warning of potential disaster risks, which not only improves the efficiency of disaster prevention and mitigation, but also provides a more scientific basis for emergency decision-making.

[0175] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0176] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A method for evaluating the working performance of earth-rock dams based on multi-source heterogeneous data input, characterized in that: The method for evaluating the working performance of an earth-rock dam based on multi-source heterogeneous data input comprises: Extract key features of image and text data from existing earth-rock dam design data to obtain a comprehensive multi-source heterogeneous data set; the comprehensive multi-source heterogeneous data set includes a dam deformation stability data set, a dam slope stability data set, a dam seepage data set, and a dam seismic safety data set; the comprehensive multi-source heterogeneous data set is used to train a TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is used to predict the working performance of earth-rock dams; Based on the One-hot encoder, the comprehensive multi-source heterogeneous dataset is subjected to cross-channel fusion processing to obtain an input dataset and an output label dataset of the earth-rock dam image-text heterogeneous high-dimensional feature data; According to the input data set and output label data set of heterogeneous high-dimensional feature data of earth-rock dam images and texts, the TTAO-EDconv-LSTM-Attention prediction model is trained to obtain a trained TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is a deep learning model that introduces TTAO mechanism, EDconv and Attention mechanism on the basis of the original LSTM neural network; The image and text data of the target earth-rock dam are input into the trained TTAO-EDconv-LSTM-Attention prediction model to obtain the prediction results of the target earth-rock dam; the prediction results include deformation stability, dam slope stability, seepage stability and seismic safety.

2. The method for evaluating the working performance of earth-rock dams based on multi-source heterogeneous data input according to claim 1 is characterized in that: After obtaining the prediction result of the target earth-rock dam, the method further includes: The prediction results are transformed into probability distribution based on the Softmax multi-index coupling evaluator to obtain the dam safety assessment result; the dam safety assessment result is composed of the accuracy, precision, recall rate and F1 score calculated by the confusion matrix.

3. The method for evaluating the working performance of earth-rock dam based on multi-source heterogeneous data input according to claim 2 is characterized in that: The prediction results are converted into probability distribution based on the Softmax multi-index coupling evaluator, specifically including: According to the formula Performing probability distribution conversion on the prediction result; Among them, y i ∈ [ y1,y2,y3,y 4] ,y i is the predicted value of the i-th dimension, y1 is the predicted value of deformation stability, y2 is the predicted value of dam slope stability, y3 is the predicted value of seepage stability, y4 is the predicted value of seismic safety, P(y i ) is the predicted probability corresponding to the i-th dimension, and j represents the predicted dimension.

4. The method for evaluating the working performance of earth-rock dams based on multi-source heterogeneous data input according to claim 1 is characterized in that: The calculation formula for the accuracy of the confusion matrix calculation in the dam safety assessment result is: The calculation formula for the accuracy of the dam safety assessment results is: The calculation formula for the recall rate in the dam safety assessment results is: The calculation formula of F1-Score in the dam safety assessment result is: In the formula, TP is a true positive example, FP is a false positive example, FN is a false negative example, TN is a true negative example, Recall is the recall rate, Precision is the accuracy rate, and F1-Score is the harmonic average of precision and recall rate.

5. The method for evaluating the working performance of earth-rock dam based on multi-source heterogeneous data input according to claim 1 is characterized in that: The key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including: Construct a dam deformation stability dataset, including: Collect multi-source data of earth-rock dams and extract geometric parameters of the earth-rock dams; the geometric parameters include dam body shape, size, slope, soil type, permeability coefficient, density, cohesion, internal friction angle, and water level change; According to the extracted geometric parameters, the finite element model of the earth-rock dam is established based on the finite element calculation software; Performing deformation analysis on the finite element model of the earth-rock dam to obtain the maximum tensile stress, cracking conditions and settlement of the earth-rock dam during completion under different conditions; A deformation stability dataset is constructed based on the maximum tensile stress, cracking and settlement of the earth-rock dam under different conditions.

6. The method for evaluating the working performance of earth-rock dams based on multi-source heterogeneous data input according to claim 5 is characterized in that: The key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including: Construct a dam slope stability dataset, including: Based on the limit equilibrium method, the shape and position of the sliding surface are defined; According to the formula The dam slope stability analysis is performed on the earth-rock dam finite element model to obtain a dam slope stability data set.

7. The method for evaluating the working performance of earth-rock dam based on multi-source heterogeneous data input according to claim 6 is characterized in that: The key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including: Construct a dam body seepage data set, including: Performing seepage analysis on the finite element model of the earth-rock dam to generate key indicators of seepage volume, pressure head and hydraulic gradient; A dam seepage data set is constructed based on the generated key indicators of dam profile, water level, permeability coefficient, seepage rate, pressure head and hydraulic gradient.

8. The method for evaluating the working performance of earth-rock dams based on multi-source heterogeneous data input according to claim 7 is characterized in that: The key features of image and text data of existing earth-rock dam design data are extracted to obtain a comprehensive multi-source heterogeneous data set, including: Construct a dam seismic safety data set, including: Performing dynamic time history analysis on the finite element model of the earth-rock dam to obtain key indicators of dynamic safety factor, liquefaction area and residual deformation; Based on the key indicators of dynamic safety factor, liquefaction area and residual deformation, a dam seismic safety dataset is constructed.

9. An earth-rock dam working performance evaluation system based on multi-source heterogeneous data input, characterized in that: include: A comprehensive multi-source heterogeneous data set construction module is used to extract key features of image and text data from existing earth-rock dam design data to obtain a comprehensive multi-source heterogeneous data set; the comprehensive multi-source heterogeneous data set includes a dam deformation stability data set, a dam slope stability data set, a dam seepage data set, and a dam seismic safety data set; the comprehensive multi-source heterogeneous data set is used to train a TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is used to predict the working performance of earth-rock dams; A fusion processing module is used to perform cross-channel fusion processing on the comprehensive multi-source heterogeneous data set based on a One-hot encoder to obtain an input data set and an output label data set of earth-rock dam image-text heterogeneous high-dimensional feature data; A model training module is used to train a TTAO-EDconv-LSTM-Attention prediction model according to an input data set and an output label data set of heterogeneous high-dimensional feature data of earth-rock dam images and texts, so as to obtain a trained TTAO-EDconv-LSTM-Attention prediction model; the TTAO-EDconv-LSTM-Attention prediction model is a deep learning model that introduces TTAO mechanism, EDconv and Attention mechanism on the basis of the original LSTM neural network; The prediction module is used to input the image and text data of the target earth-rock dam into the trained TTAO-EDconv-LSTM-Attention prediction model to obtain the prediction results of the target earth-rock dam; the prediction results include deformation stability, dam slope stability, seepage stability and seismic safety.

10. The earth-rock dam working performance evaluation system based on multi-source heterogeneous data input according to claim 9 is characterized in that: Also includes: A conversion module, used for converting the prediction result into a probability distribution based on a Softmax multi-index coupling evaluator to obtain a dam safety assessment result; The dam safety assessment result is composed of the accuracy, precision, recall and F1 score calculated by the confusion matrix.

Images