Intelligent grouting effect comprehensive evaluation system and method

Abstract

The invention discloses an intelligent grouting effect comprehensive evaluation system and method, and the system comprises an obtaining module which is used for obtaining water pressure test data, sound wave velocity detection data and load test data corresponding to a to-be-evaluated dam, and constructing an original data set; the preprocessing module is used for preprocessing the original data set to obtain a preprocessed data set; the feature extraction module is used for performing multi-scale feature extraction on the preprocessed data set to obtain a time scale feature, a space scale feature and a physical scale feature; the fusion module is used for fusing the time scale feature, the spatial scale feature and the physical scale feature based on a space-time attention fusion mechanism to obtain a fusion feature vector; the evaluation module is used for inputting the fusion feature vector into a grouting effect prediction model for grouting effect prediction and outputting a grouting effect evaluation result; and outputting a grouting effect evaluation result by using the grouting effect prediction model, thereby providing scientific support for quality evaluation of the dam grouting project.

Description

Technical Field

[0001] This invention relates to the field of grouting effect evaluation technology, and in particular to an intelligent grouting effect comprehensive evaluation system and method. Background Technology

[0002] Grouting technology is a core, concealed engineering technology for seepage prevention and reinforcement of major infrastructure such as dams, and its construction quality directly determines the long-term stability and safety of the project. Therefore, accurate evaluation of grouting effects is a key aspect of project quality control.

[0003] However, in actual engineering projects, the evaluation of dam grouting effects has long faced many bottlenecks, and traditional evaluation models are unable to meet the needs of precise and intelligent engineering: Evaluations often rely on the experience and judgment of technical personnel or traditional statistical analysis methods, which are highly subjective and have limited ability to handle complex data relationships. Although some studies have introduced machine learning models, the prediction accuracy and generalization ability of these models are difficult to meet the needs of different engineering scenarios due to the low quality of input features.

[0004] Therefore, there is an urgent need for an intelligent grouting effect comprehensive evaluation system and method to solve the above problems. Summary of the Invention

[0005] The present invention aims to at least partially solve one of the technical problems in the aforementioned technologies. Therefore, a first aspect of the present invention aims to propose an intelligent grouting effect comprehensive evaluation system that utilizes a grouting effect prediction model to rapidly output grouting effect evaluation results, providing scientific support for the quality assessment of dam grouting projects.

[0006] The second objective of this invention is to propose a comprehensive evaluation method for intelligent grouting effects.

[0007] To achieve the above objectives, a first aspect of the present invention provides an intelligent grouting effect comprehensive evaluation system, comprising: The acquisition module is used to acquire the pressure water test data, acoustic wave velocity detection data and load test data corresponding to the dam to be evaluated, and to construct the original dataset; The preprocessing module preprocesses the original dataset to obtain the preprocessed dataset; The feature extraction module is used to perform multi-scale feature extraction on the preprocessed dataset to obtain temporal scale features, spatial scale features and physical scale features; The fusion module is used to fuse the temporal scale features, spatial scale features and physical scale features based on the spatiotemporal attention fusion mechanism to obtain a fused feature vector; The evaluation module is used to input the fused feature vector into the grouting effect prediction model to predict the grouting effect and output the grouting effect evaluation result.

[0008] Preferably, the acquisition module includes: The first acquisition submodule is used to acquire the equipment required for the pressure test. The dam to be evaluated is drilled according to the design hole position and depth. After cleaning the rock powder in the hole, the double plug water-stopping device is lowered into the target test section and expanded and sealed. The three-stage pressurization method is used to pressurize step by step, and the pressure value, instantaneous flow rate, stable flow rate, test section length and ambient temperature are recorded. After depressurization step by step, the hole is sealed to obtain the pressure test data. The second acquisition submodule is used to acquire the equipment required for acoustic wave velocity detection. For the dam to be evaluated, a single-hole detection or cross-hole detection scheme is selected. The detection hole is cleaned and the hole wall is treated. Clean water is injected or couplant is filled. The detection point is set. The transducer is lowered and its axis is aligned and it is close to the hole wall. The acoustic wave signal is emitted and the measurement is repeated at each detection point. The depth, acoustic wave propagation time, longitudinal wave velocity and waveform data are recorded to obtain the acoustic wave velocity detection data. The third acquisition submodule is used to acquire the equipment required for the load test. A test pit is excavated in the undisturbed rock mass outside the grouting zone of the dam to be evaluated. The bearing plate is placed horizontally on the surface of the rock mass and a thin layer of sand is laid. Loading devices, reaction devices and displacement sensors are installed. The slow sustained load method is used to load the load step by step according to the design level. The load value, cumulative displacement, stable displacement and loading time are recorded. When a sudden increase in displacement or failure of the bearing plate occurs, the failure load is recorded. The load is unloaded step by step and the residual displacement is recorded to obtain the load test data.

[0009] Preferably, the preprocessing module includes: The first preprocessing submodule is used to process the peak outliers in the pressure water test data based on median filtering and the 3σ criterion, while retaining the data from the stable seepage stage. The second preprocessing submodule is used to identify discrete outliers in acoustic wave velocity data based on Gaussian mixture model clustering, and to fill out the outliers using K-nearest neighbor interpolation. The third preprocessing submodule is used to identify the elastic stage of the load test through piecewise linear fitting and to eliminate abnormal points of plastic deformation. The determination submodule is used to preserve the original scale of time series / 3D data, perform Z-score standardization on scalar features, and perform Min-Max normalization on evaluation metrics to map them to the [0,1] interval, thus obtaining the preprocessed dataset.

[0010] Preferably, the feature extraction module includes: The first feature extraction submodule is used to extract time-scale features based on water pressure test data. The time-scale features include pressure rise rate, flow stabilization time, pressure-flow hysteresis coefficient, and steady-state flow fluctuation rate. The second feature extraction submodule is used to extract spatial scale features based on the 3D wave velocity volume. The spatial scale features include wave velocity spatial variance, wave velocity gradient, crack filling rate spatial entropy, and the proportion of wave velocity anomaly regions. The third feature extraction submodule is used to extract physical scale features based on the steady-state data of pressurized water, acoustic statistics and load elastic modulus. The physical scale features include logarithmic transformation value of permeability, hydraulic conductivity, effective porosity of grouting, elastic modulus and compressive strength.

[0011] Preferably, the fusion module includes: The first fusion submodule is used to input the original time series data of the water pressure test into the time attention submodule, learn the time dependency through 1D convolutional layer and gated recurrent unit, calculate the attention weight, and output a 128-dimensional time-weighted feature vector. The second fusion submodule is used to input the 3D wave velocity volume into the spatial attention submodule, extract local correlations through the 3D convolutional layer, generate a spatial attention weight matrix, and output a 128-dimensional spatial weighted feature vector. The third fusion submodule is used to extend the physical scale feature vector to 128 dimensions through MLP, concatenate it with the time-weighted feature vector and the spatial-weighted feature vector, input it into the attention layer to calculate the interaction weights between features, fuse them through matrix multiplication, and output a fusion feature vector with a dimension of 64.

[0012] A preferred method for constructing a grouting effect prediction model includes: Three parallel subtasks were constructed: seepage prevention performance prediction subtask, density prediction subtask, and rock mass strength prediction subtask. For the output of the dual-base learner for each subtask, a forward cloud generator is used to generate a set of cloud droplets corresponding to each preliminary prediction value. The cloud droplet set is then fitted by a reverse cloud generator to obtain the membership cloud model for each subtask. An improved AHP-cloud scaling method is used to construct a judgment matrix, and the eigenvectors are calculated and normalized using the square root method. After consistency verification, the model is transformed into a weighted cloud model. The membership cloud model and weight cloud model of each subtask are weighted twice to obtain the grouting effect prediction model. A five-level evaluation cloud model is preset. The matching degree of each level of the cloud model is calculated based on the cloud similarity calculation formula. The level with the highest matching degree is taken as the evaluation result of the grouting effect.

[0013] Preferably, three parallel subtasks are constructed: a seepage prevention performance prediction subtask, a density prediction subtask, and a rock mass strength prediction subtask, including: The subtask of seepage prevention performance prediction uses a combination of random forest and gradient boosting tree as a dual-base learner, inputting a fused feature vector and outputting a preliminary prediction of seepage prevention performance. The density prediction subtask uses a combination of AdaBoost and improved ANN dual-base learners, and outputs preliminary density prediction values ​​through stacked generalization fusion. The rock mass strength prediction subtask uses a combination of XGBoost and LightGBM dual-base learners, employing a simple averaging strategy for fusion, to output preliminary rock mass strength predictions.

[0014] Preferably, the method for constructing the grouting effect prediction model further includes: The variables are optimized based on learner hyperparameters and multi-task weights; the multi-task weights satisfy the condition that the sum of the seepage prevention performance weight, the density weight, and the rock mass strength weight is 1. Construct a fitness function, which uses the average cloud similarity of the validation set as the core indicator; By setting the initial temperature and cooling coefficient, generating neighborhood solutions, calculating fitness differences, executing acceptance criteria and cooling iterations, the optimal combination of hyperparameters and multi-task weights are output.

[0015] Preferably, it also includes: a visualization module, used to map the grouting effect evaluation results onto the dam BIM model as a 3D cloud map using color gradient, generate an evaluation report including feature contribution analysis, highlight abnormal areas with cloud similarity less than a preset threshold, and mark the priority for re-inspection.

[0016] To achieve the above objectives, a second aspect of the present invention proposes a comprehensive evaluation method for intelligent grouting effects, comprising: Obtain the pressure water test data, acoustic wave velocity detection data and load test data corresponding to the dam to be evaluated, and construct the original dataset; The original dataset is preprocessed to obtain the preprocessed dataset; Multi-scale feature extraction is performed on the preprocessed dataset to obtain temporal scale features, spatial scale features, and physical scale features; The temporal scale features, spatial scale features, and physical scale features are fused based on the spatiotemporal attention fusion mechanism to obtain a fused feature vector; The fused feature vector is input into the grouting effect prediction model to predict the grouting effect and output the grouting effect evaluation result.

[0017] This invention provides an intelligent grouting effect comprehensive evaluation system and method. It constructs an original dataset by acquiring data from water pressure tests, acoustic wave velocity detection, and load tests, reflecting the dam grouting situation from multiple dimensions and providing a rich and comprehensive data foundation for accurate evaluation. A preprocessing module processes the original dataset, effectively removing noise, missing values, and other interference factors, improving data quality and providing a reliable basis for subsequent analysis. A multi-scale feature extraction module extracts features from temporal, spatial, and physical scales, deeply mining the information behind the data and more comprehensively characterizing the grouting effect-related features. A spatiotemporal attention fusion mechanism is used to fuse features at different scales, fully considering the spatiotemporal correlation and importance between features, generating a more representative fused feature vector and improving the accuracy of the evaluation. Analyzing the fused feature vector using a grouting effect prediction model allows for rapid and accurate output of grouting effect evaluation results, providing scientific support for the quality assessment and subsequent decision-making of dam grouting projects.

[0018] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a block diagram of an intelligent grouting effect comprehensive evaluation system according to an embodiment of the present invention; Figure 2 This is a block diagram of an acquisition module according to an embodiment of the present invention; Figure 3 This is a block diagram of a preprocessing module according to an embodiment of the present invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] Example 1: As Figure 1 As shown, a comprehensive evaluation system for intelligent grouting effect includes: The acquisition module is used to acquire the pressure water test data, acoustic wave velocity detection data and load test data corresponding to the dam to be evaluated, and to construct the original dataset; The preprocessing module preprocesses the original dataset to obtain the preprocessed dataset; The feature extraction module is used to perform multi-scale feature extraction on the preprocessed dataset to obtain temporal scale features, spatial scale features and physical scale features; The fusion module is used to fuse the temporal scale features, spatial scale features and physical scale features based on the spatiotemporal attention fusion mechanism to obtain a fused feature vector; The evaluation module is used to input the fused feature vector into the grouting effect prediction model to predict the grouting effect and output the grouting effect evaluation result.

[0023] The working principle and beneficial effects of the above technical solution are as follows: By acquiring data from water pressure tests, acoustic wave velocity detection, and load tests to construct an original dataset, the dam grouting situation can be reflected from multiple dimensions, providing a rich and comprehensive data foundation for accurate evaluation; the preprocessing module processes the original dataset, effectively removing noise, missing values, and other interference factors, improving data quality, and providing a reliable basis for subsequent analysis; the multi-scale feature extraction module extracts features from time, space, and physical scales, enabling in-depth mining of the information behind the data and more comprehensively characterizing the grouting effect-related features; the fusion of features at different scales based on the spatiotemporal attention fusion mechanism can fully consider the spatiotemporal correlation and importance between features, generating a more representative fused feature vector and improving the accuracy of the evaluation; the analysis of the fused feature vector using the grouting effect prediction model can quickly and accurately output the grouting effect evaluation results, providing scientific support for the quality assessment and subsequent decision-making of dam grouting projects.

[0024] Example 2: Figure 2 As shown, the acquisition module includes: The first acquisition submodule is used to acquire the equipment required for the pressure test. The dam to be evaluated is drilled according to the design hole position and depth. After cleaning the rock powder in the hole, the double plug water-stopping device is lowered into the target test section and expanded and sealed. The three-stage pressurization method is used to pressurize step by step, and the pressure value, instantaneous flow rate, stable flow rate, test section length and ambient temperature are recorded. After depressurization step by step, the hole is sealed to obtain the pressure test data. The second acquisition submodule is used to acquire the equipment required for acoustic wave velocity detection. For the dam to be evaluated, a single-hole detection or cross-hole detection scheme is selected. The detection hole is cleaned and the hole wall is treated. Clean water is injected or couplant is filled. The detection point is set. The transducer is lowered and its axis is aligned and it is close to the hole wall. The acoustic wave signal is emitted and the measurement is repeated at each detection point. The depth, acoustic wave propagation time, longitudinal wave velocity and waveform data are recorded to obtain the acoustic wave velocity detection data. The third acquisition submodule is used to acquire the equipment required for the load test. A test pit is excavated in the undisturbed rock mass outside the grouting zone of the dam to be evaluated. The bearing plate is placed horizontally on the surface of the rock mass and a thin layer of sand is laid. Loading devices, reaction devices and displacement sensors are installed. The slow sustained load method is used to load the load step by step according to the design level. The load value, cumulative displacement, stable displacement and loading time are recorded. When a sudden increase in displacement or failure of the bearing plate occurs, the failure load is recorded. The load is unloaded step by step and the residual displacement is recorded to obtain the load test data.

[0025] In this embodiment, a three-stage pressurization method is used to gradually increase the pressure. When the pressure is stable for 30 minutes and the flow rate fluctuation is ≤5%, the pressure value, instantaneous flow rate, stable flow rate, test section length and ambient temperature are recorded. After gradually decreasing the pressure, the hole is sealed to obtain the water pressure test data.

[0026] In this embodiment, the slow sustained load method is used to load the load step by step according to the design level. After each load level stabilizes until the displacement increment is ≤0.1mm within 1 hour, the load value, cumulative displacement, stable displacement and loading time are recorded. When a sudden increase in displacement or failure of the bearing plate occurs, the failure load is recorded. The load is unloaded step by step and the residual displacement is recorded to obtain the load test data.

[0027] The working principle and beneficial effects of the above technical solution are as follows: By conducting water pressure tests, acoustic wave velocity detection, and load tests through three acquisition sub-modules, data is obtained from different physical characteristics and engineering indicators, comprehensively reflecting the performance status of the grouting body and surrounding rock mass of the dam under evaluation, providing multi-dimensional data support for accurately assessing the grouting effect. Each sub-module specifies detailed test operation procedures, including equipment acquisition, preliminary preparation, test implementation, and data recording, ensuring the standardization and consistency of the data acquisition process and improving the reliability and comparability of the data. Clear requirements are set for key operational details during the test, such as the sealing of the double-plug water-stopping device, the alignment and close contact of the transducer axis with the borehole wall, and the horizontal placement of the pressure plate, reducing the impact of human factors and operational errors on the data and helping to obtain accurate and effective test data. Recording the destructive load and residual displacement in the load test allows understanding of the mechanical response and failure characteristics of the rock mass outside the grouting zone under load, further providing important information for assessing the overall stability of the dam and the grouting effect.

[0028] Example 3: As Figure 3 As shown, the preprocessing module includes: The first preprocessing submodule is used to process the peak outliers in the pressure water test data based on median filtering and the 3σ criterion, while retaining the data from the stable seepage stage. The second preprocessing submodule is used to identify discrete outliers in acoustic wave velocity data based on Gaussian mixture model clustering, and to fill out the outliers using K-nearest neighbor interpolation. The third preprocessing submodule is used to identify the elastic stage of the load test through piecewise linear fitting and to eliminate abnormal points of plastic deformation. The determination submodule is used to preserve the original scale of time series / 3D data, perform Z-score standardization on scalar features, and perform Min-Max normalization on evaluation metrics to map them to the [0,1] interval, thus obtaining the preprocessed dataset.

[0029] In this embodiment, K-Nearest Neighbors refers to the KNN algorithm.

[0030] In this embodiment, the Gaussian mixture model is the Gaussian probability density function (GMM).

[0031] In this embodiment, Min-Max refers to extreme value standardization or deviation standardization.

[0032] The working principle and beneficial effects of the above technical solution are as follows: Each preprocessing submodule uses appropriate methods to handle outliers for different types of test data. For example, median filtering and the 3σ criterion are used to handle peak outliers in water pressure test data; Gaussian mixture model clustering is used to identify discrete outliers in acoustic velocity data; and piecewise linear fitting is used to remove outliers in plastic deformation of load tests, effectively improving data accuracy and reliability. Outliers in acoustic velocity data are filled using K-nearest neighbor interpolation to avoid affecting subsequent analysis due to missing data and ensure data integrity. Piecewise linear fitting identifies the elastic stage of load tests, which helps to extract key features reflecting the true performance of the dam and provides support for accurate evaluation of grouting effects. The determination submodule uses different standardization and normalization methods according to data type, retaining the original scale for time series / 3D data, performing Z-score standardization on scalar features, and performing Min-Max normalization on evaluation indicators, so that the data has a uniform and standardized scale, which is convenient for model processing and analysis.

[0033] Example 4: Feature extraction module, including: The first feature extraction submodule is used to extract time-scale features based on water pressure test data. The time-scale features include pressure rise rate, flow stabilization time, pressure-flow hysteresis coefficient, and steady-state flow fluctuation rate. The second feature extraction submodule is used to extract spatial scale features based on the 3D wave velocity volume. The spatial scale features include wave velocity spatial variance, wave velocity gradient, crack filling rate spatial entropy, and the proportion of wave velocity anomaly regions. The third feature extraction submodule is used to extract physical scale features based on the steady-state data of pressurized water, acoustic statistics and load elastic modulus. The physical scale features include logarithmic transformation value of permeability, hydraulic conductivity, effective porosity of grouting, elastic modulus and compressive strength.

[0034] In this embodiment, the time scale characteristics include: seepage stability characteristics: pressure rise rate, flow stabilization time, pressure-flow hysteresis coefficient (time difference between pressure peak and flow peak); seepage law characteristics: logarithmic fitting coefficient of the pressure test curve, steady-state flow fluctuation rate.

[0035] In this embodiment, the spatial scale characteristics are as follows: input: time series data of water pressure test; output characteristics: pressure rise rate, flow stabilization time, pressure-flow hysteresis coefficient, and steady-state flow fluctuation rate.

[0036] In this embodiment, the physical scale features are as follows: input: 3D wave velocity volume; output features: wave velocity spatial variance, wave velocity gradient, crack filling rate spatial entropy, and wave velocity anomaly region proportion.

[0037] Intrinsic characteristics of permeability: Input: steady-state data of pressurized water, acoustic statistics, load elastic modulus; Output characteristics: logarithmic transformation value of permeability, hydraulic conductivity, effective porosity of grouting, elastic modulus, compressive strength.

[0038] The working principle and beneficial effects of the above technical solution are as follows: Feature extraction from experimental data across three different scales—time, space, and physical—comprehensively covers the various aspects of information required for dam grouting effect evaluation, providing rich and multi-dimensional feature basis for accurate assessment. Extracted features, such as pressure rise rate, wave velocity spatial variance, and logarithmic transformation value of permeability, can deeply reflect the essential characteristics of pressure water test, acoustic wave velocity detection, and load test data, helping to uncover key information about dam grouting effect hidden behind the data. These features are closely related to actual dam engineering, such as flow stabilization time and effective porosity of grout, and can be directly applied to the evaluation of dam grouting effect, quality monitoring, and subsequent engineering decisions, demonstrating strong practicality and relevance. The extracted features provide suitable input for subsequent data analysis, model construction, and evaluation, which is conducive to improving analysis efficiency and the accuracy of evaluation results, and promoting the development of dam engineering-related research and practice.

[0039] Example 5: Fusion module, including: The first fusion submodule is used to input the original time series data of the water pressure test into the time attention submodule, learn the time dependency through 1D convolutional layer and gated recurrent unit, calculate the attention weight, and output a 128-dimensional time-weighted feature vector. The second fusion submodule is used to input the 3D wave velocity volume into the spatial attention submodule, extract local correlations through the 3D convolutional layer, generate a spatial attention weight matrix, and output a 128-dimensional spatial weighted feature vector. The third fusion submodule is used to extend the physical scale feature vector to 128 dimensions through MLP, concatenate it with the time-weighted feature vector and the spatial-weighted feature vector, input it into the attention layer to calculate the interaction weights between features, fuse them through matrix multiplication, and output a fusion feature vector with a dimension of 64.

[0040] In this embodiment, the first fusion submodule receives the following inputs: a 1D convolutional layer extracts local patterns; a GRU layer learns long-term dependencies; and attention weights. ; The attention weights are for time step t; This is the transpose of the attention weight vector; It is the hyperbolic tangent activation function; Here is the weight matrix of the GRU layer; Let be the hidden state vector of the GRU at time step t; The bias term is used; the output is a 128-dimensional time-weighted feature vector.

[0041] In this embodiment, the second fusion submodule receives the following inputs: a 3D wave velocity volume; the mechanism involves a 3D convolutional layer extracting spatial features; and spatial attention weights are applied. ; The spatial attention weights at position (i,j,k) in the 3D wave velocity volume; It is the sigmoid activation function; The output features of the 3D convolutional layer at position (i,j,k); This is the spatial attention weight matrix; For spatial attention bias term; Output: 128-dimensional spatial weighted feature vector.

[0042] In this embodiment, the third fusion submodule takes the following inputs: a time-weighted feature vector (128-dimensional); a spatially weighted feature vector (128-dimensional); and a physical scale feature vector (10-dimensional scalar concatenation) → expanded to 128 dimensions via MLP; the mechanism involves concatenating three types of features (384-dimensional); and a self-attention layer. ; For query matrix; The key matrix; It is a value matrix; is the scaling factor; output: a 64-dimensional fused feature vector.

[0043] The working principle and beneficial effects of the above technical solution are as follows: The first and second fusion submodules utilize the temporal attention submodule and spatial attention submodule respectively, combined with convolutional layers and recurrent units, to accurately capture the temporal dependencies of the original time-series data of the water pressure test and the spatial local correlations of the 3D wave velocity body, outputting weighted feature vectors and fully exploring the inherent characteristics of different types of data; through the attention mechanism, reasonable weights are assigned to different features, highlighting key information and enhancing the expressive power of the feature vectors on data features, enabling the model to focus more on important features and improve the accuracy of subsequent analysis; the third fusion submodule uniformly expands feature vectors of different scales to the same dimension and then splices them, and then uses the self-attention layer to calculate and fuse the interaction weights, realizing the effective fusion of multi-source features at the temporal, spatial and physical scales, breaking down the barriers between data and providing more comprehensive and integrated information; finally, the output is a 64-dimensional fusion feature vector, which reduces the feature dimension while retaining key information, reducing the amount of computation and storage requirements, improving the efficiency of model training and inference, and also helping to avoid the curse of dimensionality problem.

[0044] Example 6: A method for constructing a grouting effect prediction model, including: Three parallel subtasks were constructed: seepage prevention performance prediction subtask, density prediction subtask, and rock mass strength prediction subtask. For the output of the dual-base learner for each subtask, a forward cloud generator is used to generate a set of cloud droplets corresponding to each preliminary prediction value. The cloud droplet set is then fitted by a reverse cloud generator to obtain the membership cloud model for each subtask. An improved AHP-cloud scaling method is used to construct a judgment matrix, and the eigenvectors are calculated and normalized using the square root method. After consistency verification, the model is transformed into a weighted cloud model. The membership cloud model and weight cloud model of each subtask are weighted twice to obtain the grouting effect prediction model. A five-level evaluation cloud model is preset. The matching degree of each level of the cloud model is calculated based on the cloud similarity calculation formula. The level with the highest matching degree is taken as the evaluation result of the grouting effect.

[0045] In this embodiment, for the output of the dual-base learner for each subtask, a set of cloud droplets corresponding to each preliminary prediction value is generated based on the forward cloud generator. The cloud droplet set is then fitted using the inverse cloud generator to obtain the membership cloud model for each subtask, including: In this embodiment, for the output of the dual-base learner of each subtask, a forward cloud generator is used to generate a set of cloud droplets, wherein the predicted values ​​of the subtasks are... Generate 50 cloud droplets: ; These are preliminary predictions for the subtask; ( ) represents a normal distribution; Entropy; It is hyperentropy; Fitting cloud model : ; ; ; This is the expected value; The number of cloud droplets; For the k-th cloud droplet value; calculate the multi-task dynamic weights: AHP-cloud scaling method to construct the judgment matrix (scaling: C(3,0.2,0.03)=important, C(5,0.33,0.05)=significantly important); weighted cloud model: ; Let be the expected weight of subtask j; This is the chain multiplication operator; To determine the expected value of the element in the j-th row and k-th column of the matrix; fusion sub-task cloud model: ; This is the overall expected value; Let be the expected weight of subtask j; Let be the expected value of the membership degree of subtask j in the cloud model; ; This is the overall entropy value; Let be the entropy weight of subtask j; Let be the membership degree of subtask j, and be the entropy value of the cloud model. ; The overall hyperentropy value; Let be the hyperentropy weight of subtask j; The membership degree of subtask j is the hyperentropy value of the cloud model. Five-level cloud model: Excellent C (0.9, 0.05, 0.01), Good C (0.7, 0.06, 0.015), Average C (0.5, 0.07, 0.02), Poor C (0.3, 0.08, 0.025), Very Poor C (0.1, 0.09, 0.03); Cloud similarity: ; This is the overall expected value; This represents the expected value of the hierarchical cloud model. This is the overall entropy value; The entropy value of the hierarchical cloud model; The overall hyperentropy value; The hyperentropy value is the highest level of the Sim cloud model; the highest level is selected as the result.

[0046] The working principle and beneficial effects of the above technical solution are as follows: Three parallel sub-tasks—seepage prevention performance, density, and rock mass strength—are constructed to comprehensively evaluate the grouting effect from multiple key dimensions, making the evaluation results more comprehensive and accurate in reflecting the actual grouting situation. A membership cloud model is generated using forward and reverse cloud generators, effectively handling the uncertainty and ambiguity in grouting effect prediction, better reflecting complex actual situations, and improving the model's reliability. An improved AHP-cloud scaling method is used to construct a judgment matrix and obtain a weighted cloud model, scientifically and rationally determining the weight of each sub-task in the overall evaluation, avoiding subjective interference, and enhancing the objectivity of the evaluation. A secondary weighted average is performed between the membership cloud model and the weighted cloud model to achieve dynamic integration of sub-task evaluation results and weights, fully considering the differences in importance of each sub-task and optimizing the prediction model. A five-level evaluation grade cloud model is preset, and the grouting effect evaluation grade is determined by calculating the matching degree, making the evaluation results intuitive and easy to understand, facilitating application and decision-making in actual engineering projects.

[0047] Example 7: Construct three parallel subtasks: seepage prevention performance prediction subtask, density prediction subtask, and rock mass strength prediction subtask, including: The subtask of seepage prevention performance prediction uses a combination of random forest and gradient boosting tree as a dual-base learner, inputting a fused feature vector and outputting a preliminary prediction of seepage prevention performance. The density prediction subtask uses a combination of AdaBoost and improved ANN dual-base learners, and outputs preliminary density prediction values ​​through stacked generalization fusion. The rock mass strength prediction subtask uses a combination of XGBoost and LightGBM dual-base learners, employing a simple averaging strategy for fusion, to output preliminary rock mass strength predictions.

[0048] In this embodiment, the subtask of seepage prevention performance prediction uses a base learner combination of Random Forest (RF) and Gradient Boosting Tree (GBDT), and a fusion strategy of weighted average, with weights... ; denoted as the root mean square error of the base learner i on the validation set; The density prediction subtask uses a combination of base learners: AdaBoost and an improved ANN (2 hidden layers × 64 neurons, dropout rate 0.2); the fusion strategy is stacked generalization, using linear regression to fit the base learner outputs and outputting preliminary density predictions. Rock mass strength prediction subtask, base learner combination: XGBoost and LightGBM; Fusion strategy: simple averaging, outputting preliminary predicted values ​​of rock mass strength.

[0049] The working principle and beneficial effects of the above technical solution are as follows: Each subtask employs a combination of two base learners, with different learners possessing different learning capabilities and characteristics. For example, random forests can handle high-dimensional data and have strong anti-overfitting capabilities, while gradient boosting trees can continuously optimize the model through iterative training. The combination of the two can learn and predict seepage prevention performance from different perspectives, thereby improving the accuracy of seepage prevention performance prediction. Similarly, the combination of two base learners in other subtasks can also leverage their respective advantages to improve the accuracy of density and rock mass strength prediction; different base learners have different sensitivities to noise and outliers in the data. By combining multiple base learners, the prediction bias caused by data anomalies in a single learner can be reduced, enabling the entire model to maintain relatively stable prediction performance when facing various complex data, thus enhancing the model's robustness; the three subtasks are executed in parallel, making full use of computing resources and reducing the overall processing time. In practical applications, parallel processing can significantly improve prediction efficiency and provide timely support for engineering decisions for large-scale geotechnical engineering data. Different fusion strategies are adopted for different sub-tasks, such as the combination of two base learners in the seepage prevention performance prediction sub-task, the stacked generalization fusion in the compaction prediction sub-task, and the simple averaging strategy in the rock mass strength prediction sub-task. These diverse fusion strategies can select the most appropriate way to fuse the results of multiple base learners according to different task requirements and data characteristics, thereby further improving the prediction effect.

[0050] Example 8: Membership cloud model, including three numerical features: expectation Ex, entropy En, and hyperentropy He.

[0051] Example 9: A method for constructing a grouting effect prediction model, further comprising: The variables are optimized based on learner hyperparameters and multi-task weights; the multi-task weights satisfy the condition that the sum of the seepage prevention performance weight, the density weight, and the rock mass strength weight is 1. Construct a fitness function, which uses the average cloud similarity of the validation set as the core indicator; By setting the initial temperature and cooling coefficient, generating neighborhood solutions, calculating fitness differences, executing acceptance criteria and cooling iterations, the optimal combination of hyperparameters and multi-task weights are output.

[0052] In this embodiment, the variable definition is optimized: Base learner hyperparameters: number of trees in RF (50-200), learning rate of GBDT (0.01-0.3), number of hidden layer neurons in ANN (32-128), tree depth of XGBoost (3-10), etc. Multi-task weights: seepage prevention performance weight (w1), density weight (w2), rock mass strength weight (w3), satisfying w1+w2+w3=1.

[0053] In this embodiment, the fitness function ; To verify the average cloud similarity of the set; .

[0054] In this embodiment, initialization: initial temperature T0 = 100, cooling coefficient α = 0.95, initial solution is a randomly generated combination of hyperparameters and weights; iterative process: generating neighborhood solutions: randomly perturbing the current solution (hyperparameters adjusted ±10%, weights adjusted ±0.1 and normalized); ΔF = Fitness(new solution) - Fitness(current solution); acceptance criterion: if ΔF > 0, the new solution is directly accepted; cooling: Repeat the iterations until T < 1 or the maximum number of iterations is reached. Output: Optimal hyperparameter combination and multi-task weights.

[0055] The working principle and beneficial effects of the above technical solution are as follows: Simulated annealing algorithm, by setting the initial temperature and cooling coefficient, performs operations such as generating neighborhood solutions, calculating fitness differences, executing acceptance criteria, and iterating through cooling, enabling a global search across the entire parameter space. It allows for a certain probability of accepting poorer neighborhood solutions, avoiding getting trapped in local optima, thus increasing the chance of finding the truly optimal combination of hyperparameters and multi-task weights, leading to better model performance. The cooling iteration process of simulated annealing algorithm allows the model to extensively explore the parameter space in the early stages of the search, gradually converging to a better solution as the temperature decreases. This incremental optimization approach ensures both the comprehensiveness of the search and the attainment of relatively stable results within a reasonable timeframe, improving the efficiency and reliability of the optimization process.

[0056] Example 10: A visualization module is used to map the grouting effect evaluation results onto the dam BIM model as a 3D cloud map using color gradients, generate an evaluation report including feature contribution analysis, highlight abnormal areas with cloud similarity less than a preset threshold, and mark the priority for re-inspection.

[0057] The working principle and beneficial effects of the above technical solution are as follows: The grouting effect evaluation results are mapped onto the dam's BIM model using color gradients to form a 3D cloud map. This allows relevant personnel to intuitively understand the distribution of grouting effects in different parts of the dam from a three-dimensional spatial perspective. Color differences clearly reflect the quality of grouting, enabling a quick grasp of the overall grouting quality without complex data interpretation, thus improving information transmission efficiency. The generation of an evaluation report including feature contribution analysis helps to deeply understand the various factors affecting grouting effects and their importance. Engineers can use these analysis results to adjust grouting process and material parameters in a targeted manner, optimizing subsequent grouting work and providing a strong basis for scientific decision-making. Highlighting abnormal areas with cloud similarity below a preset threshold and marking re-inspection priorities helps staff quickly locate areas with potential grouting quality problems and rationally allocate resources for re-inspection based on these priorities. This greatly improves the efficiency of quality inspection and problem identification, enabling timely detection and resolution of potential safety hazards and ensuring the safety and stability of the dam.

[0058] To achieve the above objectives, a second aspect of the present invention proposes a comprehensive evaluation method for intelligent grouting effects, comprising: Obtain the pressure water test data, acoustic wave velocity detection data and load test data corresponding to the dam to be evaluated, and construct the original dataset; The original dataset is preprocessed to obtain the preprocessed dataset; Multi-scale feature extraction is performed on the preprocessed dataset to obtain temporal scale features, spatial scale features, and physical scale features; The temporal scale features, spatial scale features, and physical scale features are fused based on the spatiotemporal attention fusion mechanism to obtain a fused feature vector; The fused feature vector is input into the grouting effect prediction model to predict the grouting effect and output the grouting effect evaluation result.

[0059] The working principle and beneficial effects of the above technical solution are as follows: By acquiring data from water pressure tests, acoustic wave velocity detection, and load tests to construct an original dataset, the dam grouting situation can be reflected from multiple dimensions, providing a rich and comprehensive data foundation for accurate evaluation; the preprocessing module processes the original dataset, effectively removing noise, missing values, and other interference factors, improving data quality, and providing a reliable basis for subsequent analysis; the multi-scale feature extraction module extracts features from time, space, and physical scales, enabling in-depth mining of the information behind the data and more comprehensively characterizing the grouting effect-related features; the fusion of features at different scales based on the spatiotemporal attention fusion mechanism can fully consider the spatiotemporal correlation and importance between features, generating a more representative fused feature vector and improving the accuracy of the evaluation; the analysis of the fused feature vector using the grouting effect prediction model can quickly and accurately output the grouting effect evaluation results, providing scientific support for the quality assessment and subsequent decision-making of dam grouting projects.

[0060] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A comprehensive evaluation system for intelligent grouting effect, characterized in that, include: The acquisition module is used to acquire the pressure water test data, acoustic wave velocity detection data and load test data corresponding to the dam to be evaluated, and to construct the original dataset; The preprocessing module preprocesses the original dataset to obtain the preprocessed dataset; The feature extraction module is used to perform multi-scale feature extraction on the preprocessed dataset to obtain temporal scale features, spatial scale features and physical scale features; The fusion module is used to fuse the temporal scale features, spatial scale features and physical scale features based on the spatiotemporal attention fusion mechanism to obtain a fused feature vector; The evaluation module is used to input the fused feature vector into the grouting effect prediction model to predict the grouting effect and output the grouting effect evaluation result.

2. The intelligent grouting effect comprehensive evaluation system as described in claim 1, characterized in that, The acquisition module includes: The first acquisition submodule is used to acquire the equipment required for the pressure test. The dam to be evaluated is drilled according to the design hole position and depth. After cleaning the rock powder in the hole, the double plug water-stopping device is lowered into the target test section and expanded and sealed. The three-stage pressurization method is used to pressurize step by step, and the pressure value, instantaneous flow rate, stable flow rate, test section length and ambient temperature are recorded. After depressurization step by step, the hole is sealed to obtain the pressure test data. The second acquisition submodule is used to acquire the equipment required for acoustic wave velocity detection. For the dam to be evaluated, a single-hole detection or cross-hole detection scheme is selected. The detection hole is cleaned and the hole wall is treated. Clean water is injected or couplant is filled. The detection point is set. The transducer is lowered and its axis is aligned and it is close to the hole wall. The acoustic wave signal is emitted and the measurement is repeated at each detection point. The depth, acoustic wave propagation time, longitudinal wave velocity and waveform data are recorded to obtain the acoustic wave velocity detection data. The third acquisition submodule is used to acquire the equipment required for the load test. A test pit is excavated in the undisturbed rock mass outside the grouting zone of the dam to be evaluated. The bearing plate is placed horizontally on the surface of the rock mass and a thin layer of sand is laid. Loading devices, reaction devices and displacement sensors are installed. The slow sustained load method is used to load the load step by step according to the design level. The load value, cumulative displacement, stable displacement and loading time are recorded. When a sudden increase in displacement or failure of the bearing plate occurs, the failure load is recorded. The load is unloaded step by step and the residual displacement is recorded to obtain the load test data.

3. The intelligent grouting effect comprehensive evaluation system as described in claim 1, characterized in that, The preprocessing module includes: The first preprocessing submodule is used to process the peak outliers in the pressure water test data based on median filtering and the 3σ criterion, while retaining the data from the stable seepage stage. The second preprocessing submodule is used to identify discrete outliers in acoustic wave velocity data based on Gaussian mixture model clustering, and to fill out the outliers using K-nearest neighbor interpolation. The third preprocessing submodule is used to identify the elastic stage of the load test through piecewise linear fitting and to eliminate abnormal points of plastic deformation. The determination submodule is used to preserve the original scale of time series / 3D data, perform Z-score standardization on scalar features, and perform Min-Max normalization on evaluation metrics to map them to the [0,1] interval, thus obtaining the preprocessed dataset.

4. The intelligent grouting effect comprehensive evaluation system as described in claim 1, characterized in that, The feature extraction module includes: The first feature extraction submodule is used to extract time-scale features based on water pressure test data. The time-scale features include pressure rise rate, flow stabilization time, pressure-flow hysteresis coefficient, and steady-state flow fluctuation rate. The second feature extraction submodule is used to extract spatial scale features based on the 3D wave velocity volume. The spatial scale features include wave velocity spatial variance, wave velocity gradient, crack filling rate spatial entropy, and the proportion of wave velocity anomaly regions. The third feature extraction submodule is used to extract physical scale features based on the steady-state data of pressurized water, acoustic statistics and load elastic modulus. The physical scale features include logarithmic transformation value of permeability, hydraulic conductivity, effective porosity of grouting, elastic modulus and compressive strength.

5. The intelligent grouting effect comprehensive evaluation system as described in claim 1, characterized in that, The fusion module includes: The first fusion submodule is used to input the original time series data of the water pressure test into the time attention submodule, learn the time dependency through 1D convolutional layer and gated recurrent unit, calculate the attention weight, and output a 128-dimensional time-weighted feature vector. The second fusion submodule is used to input the 3D wave velocity volume into the spatial attention submodule, extract local correlations through the 3D convolutional layer, generate a spatial attention weight matrix, and output a 128-dimensional spatial weighted feature vector. The third fusion submodule is used to extend the physical scale feature vector to 128 dimensions through MLP, concatenate it with the time-weighted feature vector and the spatial-weighted feature vector, input it into the attention layer to calculate the interaction weights between features, fuse them through matrix multiplication, and output a fusion feature vector with a dimension of 64.

6. The intelligent grouting effect comprehensive evaluation system as described in claim 1, characterized in that, The method for constructing a grouting effect prediction model includes: Three parallel subtasks are constructed: seepage prevention performance prediction subtask, density prediction subtask, and rock mass strength prediction subtask. For the output of the dual-base learner for each subtask, a forward cloud generator is used to generate a set of cloud droplets corresponding to each preliminary prediction value. The cloud droplet set is then fitted by a reverse cloud generator to obtain the membership cloud model for each subtask. An improved AHP-cloud scaling method is used to construct a judgment matrix, and the eigenvectors are calculated and normalized using the square root method. After consistency verification, the model is transformed into a weighted cloud model. The membership cloud model and weight cloud model of each subtask are weighted twice to obtain the grouting effect prediction model. A five-level evaluation cloud model is preset. The matching degree of each level of the cloud model is calculated based on the cloud similarity calculation formula. The level with the highest matching degree is taken as the evaluation result of the grouting effect.

7. The intelligent grouting effect comprehensive evaluation system as described in claim 6, characterized in that, Three parallel subtasks are constructed: a seepage prevention performance prediction subtask, a density prediction subtask, and a rock mass strength prediction subtask, including: The subtask of seepage prevention performance prediction uses a combination of random forest and gradient boosting tree as a dual-base learner, inputting a fused feature vector and outputting a preliminary prediction of seepage prevention performance. The density prediction subtask uses a combination of AdaBoost and improved ANN dual-base learners, and outputs preliminary density prediction values ​​through stacked generalization fusion. The rock mass strength prediction subtask uses a combination of XGBoost and LightGBM dual-base learners, employing a simple averaging strategy for fusion, to output preliminary rock mass strength predictions.

8. The intelligent grouting effect comprehensive evaluation system as described in claim 6, characterized in that, The method for constructing a grouting effect prediction model also includes: The variables are optimized based on learner hyperparameters and multi-task weights; the multi-task weights satisfy the condition that the sum of the seepage prevention performance weight, the density weight, and the rock mass strength weight is 1. Construct a fitness function, which uses the average cloud similarity of the validation set as the core indicator; By setting the initial temperature and cooling coefficient, generating neighborhood solutions, calculating fitness differences, executing acceptance criteria and cooling iterations, the optimal combination of hyperparameters and multi-task weights are output.

9. The intelligent grouting effect comprehensive evaluation system as described in claim 1, characterized in that, Also includes: The visualization module is used to map the grouting effect evaluation results onto the dam BIM model as a 3D cloud map using color gradients, generate an evaluation report including feature contribution analysis, highlight abnormal areas with cloud similarity less than a preset threshold, and mark the priority for re-inspection.

10. A method for applying the intelligent grouting effect comprehensive evaluation system as described in any one of claims 1-9, characterized in that, include: Obtain the pressure water test data, acoustic wave velocity detection data and load test data corresponding to the dam to be evaluated, and construct the original dataset; The original dataset is preprocessed to obtain the preprocessed dataset; Multi-scale feature extraction is performed on the preprocessed dataset to obtain temporal scale features, spatial scale features, and physical scale features; The temporal scale features, spatial scale features, and physical scale features are fused based on the spatiotemporal attention fusion mechanism to obtain a fused feature vector; The fused feature vector is input into the grouting effect prediction model to predict the grouting effect and output the grouting effect evaluation result.

Images