Mine goaf instability risk grade classification prediction method and system and medium

Abstract

The invention belongs to the technical field of mining engineering, and particularly discloses a mine goaf instability risk grade classification prediction method and system and a medium. The prediction method comprises the following steps: acquiring goaf related data; performing preprocessing including text digitization on the data; carrying out hyper-parameter optimization on the CNN-LSTM by adopting an improved WOA; establishing a WOA '-CNN-LSTM model, substituting the preprocessed data into the model for training, and predicting risk level classification; the risk level classification prediction is evaluated by adopting accuracy, specificity, recall rate and F1-Score indexes; and carrying out textualization processing on the risk level classification prediction to obtain an intelligent classification result. The prediction system comprises a data acquisition module, a preprocessing module, a hyper-parameter optimization module, an intelligent classification module, a classification evaluation module and a result output module. The method has the characteristics of full-process automation, high reliability, and obvious improvement of accuracy and specificity.

Description

Technical Field

[0001] This application belongs to the field of mining engineering technology, specifically relating to a method, system and medium for classifying and predicting the risk level of instability in goaf areas of mines that is fully automated, highly reliable, and significantly improves accuracy and specificity. Background Technology

[0002] Mining goaf areas are vacant areas formed on the surface or underground after mineral resource extraction. Their long-term existence easily leads to instability disasters such as roof collapse, surface subsidence, slope landslides, and pillar failure, threatening not only mine production safety but also potentially causing ecological damage and property loss. Classification and prediction of goaf instability risk levels is a core aspect of mine safety management. Scientifically assessing risk levels provides a basis for disaster prevention and engineering decisions, and is crucial for ensuring the sustainable operation of mines.

[0003] The classification and prediction of goaf instability risk levels requires comprehensive consideration of multiple factors, including geological conditions (lithology, structure), mining parameters (depth, thickness, goaf size), time effects (creep, weathering), and external loads (earthquakes, groundwater). As mining depth and complexity increase, traditional empirical assessment and numerical simulation methods are no longer sufficient to meet the demand for "precise, efficient, and dynamic" prediction of goaf instability risk.

[0004] Machine learning and artificial intelligence technologies, with their core advantages of "multi-source data fusion, nonlinear relationship mining, and autonomous iterative optimization," have become a research hotspot in the field of mine goaf risk prediction in recent years. Among existing technologies, based on differences in goaf data types and model architectures, mainstream technical solutions can be divided into traditional machine learning methods, deep learning methods, and ensemble learning methods.

[0005] Traditional machine learning methods are based on statistical learning theory and construct input-output mapping relationships by manually designing features. Examples include logistic regression (LR), support vector machines (SVM), decision trees (DT), and k-nearest neighbors (KNN). Although these models have simple structures and strong interpretability, they suffer from limitations such as only being able to capture linear relationships and having weak modeling capabilities for nonlinear factors (such as the complex connectivity of fractured networks). Furthermore, feature engineering requires manual selection and parameter tuning. Therefore, they are only suitable for small and medium-sized mines (with limited data accumulation) or preliminary risk screening.

[0006] Deep learning methods automatically learn data features through multi-layered neural networks (without manual design), thus handling unstructured data such as images and time series data. They are widely used in large-scale or intelligent mines with abundant data. Common types include Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN) / Long Short-Term Memory Networks (LSTM), and Graph Neural Networks (GNN) / Graph Convolutional Networks (GCN). However, they also have drawbacks. For example, CNNs require a large amount of labeled image data, leading to high costs; the large number of model parameters necessitates high-performance GPUs for training; and the feature extraction process is "black box," hindering understanding and trust from field personnel. RNN / LSTM training is prone to gradient vanishing / exploding, has high requirements for the integrity of time series data, and its accuracy is low due to the lack of clear time series features in short-term monitoring data. GNN / GCN has complex graph structure construction, requires extensive geological knowledge, and is prone to overfitting when the sample size is small or the graph structure is sparse. It also has high computational complexity and long training cycles for large-scale mine graph models.

[0007] Ensemble learning methods reduce the bias and variance of a single model by combining multiple base models (such as multiple decision trees or multiple CNNs), thereby improving prediction accuracy and robustness. This is particularly suitable for high-risk mines where high accuracy is required. Common ensemble learning methods include Random Forest (RF), Gradient Boosting Decision Tree (GBDT) / Extreme Gradient Boosting (XGBoost), and ensemble deep learning models (such as CNN+LSTM or RF+GBDT). However, RF suffers from high model complexity (multiple trees stacked), large memory consumption during prediction, weak ability to identify rare samples such as "extremely unstable" data, and inability to handle unstructured data such as images and time series. GBDT / XGBoost suffers from serial training processes that cannot be parallelized, long training cycles with large datasets, susceptibility to overfitting (requiring rigorous tuning of regularization parameters), and sensitivity to parameters (learning rate, tree depth, etc., require repeated adjustments). Integrating deep learning models presents challenges such as complex model architecture, high design and debugging difficulty, the need for cross-domain technologies (image, time series, statistics), extremely high training and deployment costs (requiring multimodal data annotation and high-performance computing equipment), and large model size, making it difficult to deploy on edge mining equipment. Summary of the Invention

[0008] This invention addresses the shortcomings of existing technologies by proposing a fully automated, highly reliable, and significantly improved method for classifying and predicting the risk level of instability in mining goaf areas. It also provides a system for classifying and predicting the risk level of instability in mining goaf areas, as well as a computer-readable medium.

[0009] The mine goaf instability risk level classification and prediction method of the present invention is implemented as follows: including data acquisition, preprocessing, hyperparameter optimization, intelligent classification, classification evaluation, and result output steps, the specific steps of which are as follows: A. Data Acquisition: Acquire relevant data for stability rating of goaf areas; B. Preprocessing: Perform preprocessing on the aforementioned data, including text digitization; C. Hyperparameter optimization: The improved whale optimization algorithm is used to optimize the hyperparameters of CNN-LSTM; D. Intelligent classification: Build an improved CNN-LSTM model based on the improved whale optimization algorithm, input the processed data into the aforementioned model for training, and then perform risk level classification prediction; E. Classification Assessment: Using accuracy rate Acc Specificity Spe Recall rate Rec F1 -Score The indicators are used to evaluate the aforementioned models, and the better model is selected based on the evaluation indicators. Then, the predicted risk level classification result of the better model is used as the final result. F. Output Results: The final results are processed into text to obtain the intelligent classification results.

[0010] Furthermore, the data related to the stability rating of the goaf in step A includes burial depth, exposed roof area, goaf volume, goaf dip angle, adjacent goaf conditions, rock compressive strength, geological structure, rock mass structure, and instability risk level. In step B, the aforementioned data is preprocessed by first digitizing the text, then handling missing values ​​and outliers in the goaf stability rating data and the digitized text, and finally performing data normalization.

[0011] Furthermore, the geological structures, rock mass structures, and instability risk levels are all text data. The geological structures include those without faults, those with folds but minimal fold influence, those partially cut by faults or with significant fold influence, and those penetrating the surrounding rock. The rock mass structures include intact blocky structures, layered structures, fractured structures, and loose structures. The instability risk levels include very low instability risk, low instability risk, relatively high instability risk, and very high instability risk. The adjacent goaf conditions include those with no other goafs within the affected area, goafs of average size and few in number, goafs of large size and numerous and scattered, and goafs of large size and numerous and concentrated. The text digitization processing includes: The geological structures without faults, with folds but minimal fold influence, partially cut by faults or significantly affected by folds, and with faults penetrating the surrounding rock are digitized as 1, 2, 3, and 4, respectively. The complete blocky structure, layered structure, fractured structure, and loose structure in the rock mass are digitized as 1, 2, 3, and 4, respectively. The instability risk levels are categorized as very small, low, large, and very large, and assigned numerical values ​​of 1, 2, 3, and 4 respectively. The following are the digitizations for the adjacent goaf situation: no other goaf within the affected area, goaf area of ​​average size and small number, goaf area of ​​large size and large number and scattered distribution, and goaf area of ​​large size and large number and concentrated distribution, corresponding to 1, 2, 3, and 4.

[0012] Furthermore, the specific steps of the improved whale optimization algorithm are as follows: C10. Introduce Logistic chaotic mapping to initialize the population: In the formula: i =1,2,…,N are the population individual indices; j =1,2,…,D are dimension indices; μ These are the parameters for controlling chaos. K This represents the number of preheating iterations. u Let be a random variable uniformly distributed in the interval [0,1]. For the first i Initial values ​​for each individual; for i Individuals pass through k The chaotic value after the next chaotic iteration; for i Individuals pass through k The chaotic value after -1 chaotic iterations; lb j For the first j The lower bound of a dimension; ub j For the first j The upper bound of a dimension; Indicates the first i The individual in the first j The position of the dimension; After preheating j The chaotic value corresponding to the dimension; C20. Add the Levy flight strategy to the prey search phase of the whale optimization algorithm, as shown in the following formula: C21. Random Vector Generation: C22. Calculation of scale parameters and step size: C23, Location Update: In the formula: Represents random variables u Follows a mean of 0 and a variance of The normal distribution For scale parameters; Represents random variables v Follows a mean of 0 and a variance of The normal distribution Γ is the Gamma function; Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; Indicates the location of a random individual; A and C For coefficient vectors; Indicates the first t During the nth iteration i The individual in the first j The position of the dimension; α This is an adaptive step size factor; step Let levy be the step size; ub , lb Define the upper and lower boundaries of the algorithm; C30. A nonlinear adjustment development strategy is added to the spiral contraction stage of the whale optimization algorithm, as shown in the following formula: In the formula: Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; The optimal position is known. u 1. u 2. u 3 is a random number uniformly distributed within the interval [0,1].

[0013] Furthermore, the hyperparameter optimization of CNN-LSTM using the improved whale optimization algorithm involves optimizing the number of hidden node layers, initial learning rate, and L2 regularization coefficient of the LSTM model using the improved whale optimization algorithm. The establishment of the CNN-LSTM model optimized by the improved whale optimization algorithm involves connecting the output layer of the CNN model to the input gate of the LSTM model optimized by the improved whale optimization algorithm to obtain the WOA'-CNN-LSTM model.

[0014] Furthermore, the data processed in step D is substituted into the aforementioned model for training, and then risk level classification prediction is performed. This involves dividing the data preprocessed in step B and / or historical data into training and testing sets. The training set is substituted into the WOA'-CNN-LSTM model for training, and then the testing set is substituted into the trained WOA'-CNN-LSTM model for testing and correction. Finally, the preprocessed data is substituted into the corrected WOA'-CNN-LSTM model for instability risk level classification prediction.

[0015] Furthermore, in step E, accuracy Acc for: Specificity Spe for: Recall rate Rec for:

[0016] F1 -Score for: In the formula: TP For a real example; FP This is a false positive example. TN A true counterexample; FN This is a false counterexample.

[0017] Furthermore, the process includes steps such as data acquisition, preprocessing, hyperparameter optimization, intelligent classification, classification evaluation, and result output, as detailed below: The data acquisition module is used to acquire data related to the stability rating of goaf areas; The preprocessing module is used to preprocess the aforementioned data, including text digitization processing; The hyperparameter optimization module is used to optimize the hyperparameters of CNN-LSTM using an improved whale optimization algorithm. The intelligent classification module is used to build an improved CNN-LSTM model based on the improved whale optimization algorithm. The processed data is fed into the aforementioned model for training and risk level classification prediction. The classification evaluation module is used to assess accuracy. Acc Specificity Spe Recall rate Rec F1 -Score The aforementioned models are evaluated, and the superior model is selected based on the evaluation results. Then, the predicted risk level classification result of the superior model is used as the final result. The results output module is used to process the final results into text and obtain intelligent classification results.

[0018] Furthermore, the specific steps for improving the whale optimization algorithm in the hyperparameter optimization module are as follows: C10. Introduce Logistic chaotic mapping to initialize the population: In the formula: i =1,2,…,N are the population individual indices; j =1,2,…,D are dimension indices; μ These are the parameters for controlling chaos. K This represents the number of preheating iterations. u Let be a random variable uniformly distributed in the interval [0,1]. For the first iInitial values ​​for each individual; for i Individuals pass through k The chaotic value after the next chaotic iteration; for i Individuals pass through k The chaotic value after -1 chaotic iterations; lb j For the first j The lower bound of a dimension; ub j For the first j The upper bound of a dimension; Indicates the first i The individual in the first j The position of the dimension; After preheating j The chaotic value corresponding to the dimension; C20. Add the Levy flight strategy to the prey search phase of the whale optimization algorithm, as shown in the following formula: C21. Random Vector Generation: C22. Calculation of scale parameters and step size: C23, Location Update: In the formula: Represents random variables u Follows a mean of 0 and a variance of The normal distribution For scale parameters; Represents random variables v Follows a mean of 0 and a variance of The normal distribution Γ is the Gamma function; Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; Indicates the location of a random individual; A and C For coefficient vectors; Indicates the first t During the nth iteration i The individual in the first j The position of the dimension; α This is an adaptive step size factor; step Let levy be the step size; ub , lb Define the upper and lower boundaries of the algorithm; C30. A nonlinear adjustment development strategy is added to the spiral contraction stage of the whale optimization algorithm, as shown in the following formula: In the formula: Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; The optimal position is known. u 1. u 2. u 3 is a random number uniformly distributed within the interval [0,1].

[0019] The computer-readable storage medium of the present invention is implemented as follows: it stores a computer program, which, when executed by a processor, implements the aforementioned method for classifying and predicting the risk level of instability in goaf areas of mines.

[0020] The beneficial effects of this invention are: 1. Traditional methods (such as traditional machine learning and single deep learning models) struggle to simultaneously process structured data (e.g., burial depth, mining thickness) and unstructured text data (e.g., geological structures, rock mass structures). This invention addresses this by digitizing text data, transforming various types of text data, such as geological structures (no faults → 1, fault penetration → 4) and rock mass structures (intact blocky → 1, loose → 4), into computable numerical features. These features are then fused with structured data such as burial depth and roof exposure area to form a multi-dimensional input matrix. This significantly enhances the model's ability to represent the complex environment of goaf areas, ensuring data format consistency across different mines and scenarios, improving model generalization capabilities, and avoiding feature loss or distortion caused by data type fragmentation. This lays the foundation for subsequent high-precision predictions. In particular, the preprocessing introduces "missing value / outlier handling → data normalization", which can avoid the problem of low model input quality caused by "arbitrary preprocessing steps and incomplete data cleaning" in traditional methods. At the same time, it eliminates the need for manual feature selection, reduces prediction errors caused by human judgment bias, and achieves "no human intervention" automation from data acquisition to preprocessing.

[0021] 2. Traditional deep learning models (such as CNN-LSTM) rely on manual hyperparameter tuning, which suffers from low efficiency and susceptibility to overfitting or underfitting. This invention introduces an improved Whale Optimization Algorithm (WOA'), employing Logistic chaotic mapping to initialize the population. This involves generating the initial population through chaotic iteration, increasing population diversity and distribution uniformity, and avoiding local convergence caused by traditional random initialization. Furthermore, Levy flight is incorporated during the prey search phase, utilizing its long jump characteristics to escape local optima and enhancing the ability to explore complex hyperparameter spaces. In the spiral contraction phase, nonlinear adjustment factors (such as random numbers u1, u2, u3) are introduced to dynamically adjust the development step size, preventing premature convergence and ensuring global optimality of hyperparameter optimization. Simultaneously, the WOA' algorithm automatically completes hyperparameter optimization, eliminating the need for manual tuning. This effectively shortens the model's hyperparameter tuning time and avoids performance waste caused by incomplete manual tuning, ensuring that the tuned CNN-LSTM model is always in "optimal parameter configuration." This significantly improves the model's training efficiency and stability, enhancing its performance and efficiency.

[0022] 3. Addressing the limitations of traditional models (such as traditional machine learning's ability to handle only linear relationships, the difficulty of capturing temporal dependencies with a single CNN, and the weak spatial feature extraction capability of a single LSTM) in complex goaf scenarios, this invention employs CNNs to extract spatial features: convolutional and pooling layers automatically extract local spatial features (such as fracture distribution and rock stratum dip angle) from goaf data (such as 3D point clouds and geological profiles), avoiding reliance on experience from manual feature engineering; LSTM's memory units capture long-term dependencies in goaf monitoring data (such as surface displacement time histories and microseismic event frequencies), solving the temporal modeling failure problem caused by vanishing gradients in traditional RNNs; and WOA' optimizes the connection parameters between the CNN output layer and the LSTM input gate, enhancing the efficiency of converting spatial features to temporal features. Through this cascaded fusion of CNN and LSTM, the complementary advantages of CNN and LSTM are achieved, thereby improving the adaptability to various data types.

[0023] 4. Traditional methods often rely on a single accuracy metric, which fails to comprehensively reflect the model's ability to identify "high-risk classes." This invention uses accuracy (… A cc ) to measure overall prediction accuracy, ensuring the model's basic ability to identify all risk levels; employing specificity ( S pe This measures the correct identification rate of "non-high-risk samples" to avoid misclassifying low-risk samples as high-risk samples, leading to over-governance; recall rate is used. R ecThis measures the correct identification rate of "high-risk samples" to address the problem of "missed detections" in traditional models due to the limited number of high-risk samples; it employs an F1-Score comprehensive evaluation method. R ec and S pe This balances the accuracy and recall of the model. Through the above four-dimensional evaluation system, the model can be optimized in a targeted manner (e.g., if...). R ec The accuracy is relatively low (this can be improved by adjusting the parameters of the WOA' algorithm to focus more on high-risk samples). This avoids the shortcomings of traditional models that only pursue overall accuracy while ignoring the omission of high-risk samples. It meets the core objective of "prioritizing the prevention and control of high risks" in mine safety management and makes the prediction results more credible for engineering purposes. In particular, it can effectively reduce the risk of safety accidents caused by omissions or misjudgments in high-risk mines.

[0024] 5. Traditional methods (such as physical model experiments and ensemble learning) suffer from long cycles, high costs, or deployment difficulties. This invention automates text digitization, missing / outlier handling, and normalization through pre-defined algorithms, reducing manual intervention. Furthermore, WOA' automatically searches for optimal hyperparameters, enabling the CNN-LSTM model to automatically complete training and tuning. The prediction results are directly output as risk levels through text processing, facilitating rapid decision-making by mine managers. These improvements not only achieve full-process automation, reducing labor and deployment costs, but also result in a more compact parameter scale and faster inference speed for the WOA'-optimized CNN-LSTM model, meeting the real-time prediction needs of dynamic mining scenarios.

[0025] 6. Addressing the issue of insufficient generalization ability of traditional methods under small sample and complex geological conditions, the global search capability of WOA' in this invention enables the finding of optimal hyperparameter combinations through chaotic initialization and Levy flight, even in small sample scenarios, thus avoiding overfitting. Furthermore, through the feature adaptability of CNN-LSTM, it can automatically extract key features under different geological conditions (such as lithology and structure), reducing reliance on human experience. And through multi-type data fusion, it can be compatible with both structured and unstructured data, flexibly adapting to the data acquisition conditions of different mines (e.g., in some mines lacking image data, effective prediction can still be achieved through text and numerical data). These improvements not only enhance the adaptability of this invention but also cover applications in complex scenarios.

[0026] In summary, this invention optimizes the CNN-LSTM model through WOA' and integrates multi-source data with advanced algorithms, thus significantly outperforming traditional methods in terms of data fusion, model performance, comprehensiveness of evaluation, level of automation, and adaptability to different scenarios. It effectively addresses the core requirements of "precision, efficiency, and dynamism" in predicting the instability risk of goaf areas, providing reliable technical support for mine safety management and promoting the upgrading of intelligent monitoring and early warning systems in the mining engineering field. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method for classifying and predicting the risk level of instability in goaf areas in this invention. Figure 2 This is a flowchart of the WOA' of the present invention; Figure 3 This is a bar chart showing the classification prediction results in the example. Figure 4 The example shows the prediction results of WOA-CNN-LSTM. Figure 5 The example shows the prediction results of WOA'-CNN-LSTM. In the diagram: S100 - Data acquisition, S200 - Preprocessing, S300 - Hyperparameter optimization, S310 - Introduction of Logistic chaotic mapping to initialize the population, S320 - Adding Levy flight strategy to the prey search stage of the whale optimization algorithm, S330 - Adding nonlinear adjustment development strategy to the spiral contraction stage of the whale optimization algorithm, S400 - Intelligent classification, S500 - Classification evaluation, S600 - Result output. Detailed Implementation

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this does not limit the present invention in any way. Any changes or improvements made based on the teachings of the present invention shall fall within the protection scope of the present invention.

[0029] like Figure 1 As shown, the mine goaf instability risk level classification and prediction method of the present invention includes data acquisition, preprocessing, hyperparameter optimization, intelligent classification, classification evaluation, and result output steps. The specific steps are as follows: A. Data Acquisition: Acquire relevant data for stability rating of goaf areas; B. Preprocessing: Perform preprocessing on the aforementioned data, including text digitization; C. Hyperparameter optimization: The improved whale optimization algorithm is used to optimize the hyperparameters of CNN-LSTM; D. Intelligent classification: Build an improved CNN-LSTM model based on the improved whale optimization algorithm, input the processed data into the aforementioned model for training, and then perform risk level classification prediction; E. Classification Assessment: Using accuracy rate Acc Specificity Spe Recall rate Rec F1 -Score The indicators are used to evaluate the aforementioned models. Based on the evaluation indicators (the model is optimal when the evaluation indicator is close to 1), a better model is selected, and then the predicted risk level classification result of the better model is used as the final result. F. Output Results: The final results are processed into text (i.e., the numerical results are converted into text) to obtain the intelligent classification results (i.e., the textual results are obtained).

[0030] The data related to the stability rating of the goaf in step A includes burial depth, exposed roof area, goaf volume, goaf dip angle, adjacent goaf conditions, rock compressive strength, geological structure, rock mass structure, and instability risk level. In step B, the aforementioned data is preprocessed by first digitizing the text, then handling missing values ​​and outliers in the goaf stability rating data and the digitized text, and finally performing data normalization.

[0031] The geological structures, rock mass structures, and instability risk levels are all text data. The geological structures include those without faults, those with folds but minimal fold influence, those partially cut by faults or with significant fold influence, and those penetrating the surrounding rock. The rock mass structures include intact blocky structures, layered structures, fractured structures, and loose structures. The instability risk levels include very low instability risk, low instability risk, relatively high instability risk, and very high instability risk. The adjacent goaf conditions include no other goafs within the affected area, goafs of average size and few in number, goafs of large size and numerous and scattered, and goafs of large size and numerous and concentrated. The text digitization process includes: The geological structures without faults, with folds but minimal fold influence, partially cut by faults or significantly affected by folds, and with faults penetrating the surrounding rock are digitized as 1, 2, 3, and 4, respectively. The complete blocky structure, layered structure, fractured structure, and loose structure in the rock mass are digitized as 1, 2, 3, and 4, respectively. The instability risk levels are categorized as very small, low, large, and very large, and assigned numerical values ​​of 1, 2, 3, and 4 respectively. The following are the digitizations for the adjacent goaf situation: no other goaf within the affected area, goaf area of ​​average size and small number, goaf area of ​​large size and large number and scattered distribution, and goaf area of ​​large size and large number and concentrated distribution, corresponding to 1, 2, 3, and 4.

[0032] The specific steps of the improved whale optimization algorithm are as follows: C10. Introduce Logistic chaotic mapping to initialize the population: In the formula: i =1,2,…,N are the population individual indices; j =1,2,…,D are dimension indices; μ These are the parameters for controlling chaos. K This represents the number of preheating iterations. u Let be a random variable uniformly distributed in the interval [0,1]. For the first i Initial values ​​for each individual; for i Individuals pass through k The chaotic value after the next chaotic iteration; for i Individuals pass through k The chaotic value after -1 chaotic iterations; lb j For the first j The lower bound of a dimension; ub j For the first j The upper bound of a dimension; Indicates the first i The individual in the first j The position of the dimension; After preheating j The chaotic value corresponding to the dimension; C20. Add the Levy flight strategy to the prey search phase of the whale optimization algorithm, as shown in the following formula: C21. Random Vector Generation: C22. Calculation of scale parameters and step size: C23, Location Update: In the formula: Represents random variables u Follows a mean of 0 and a variance of The normal distribution For scale parameters; Represents random variables v Follows a mean of 0 and a variance of The normal distribution Γ is the Gamma function; Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; Indicates the location of a random individual; A and C For coefficient vectors; Indicates the first t During the nth iterationi The individual in the first j The position of the dimension; α This is an adaptive step size factor; step Let levy be the step size; ub , lb Define the upper and lower boundaries of the algorithm; C30. A nonlinear adjustment development strategy is added to the spiral contraction stage of the whale optimization algorithm, as shown in the following formula: In the formula: Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; The optimal position is known. u 1. u 2. u 3 is a random number uniformly distributed within the interval [0,1].

[0033] The hyperparameter optimization of CNN-LSTM using the improved whale optimization algorithm involves optimizing the number of hidden node layers, initial learning rate, and L2 regularization coefficient of the LSTM model using the improved whale optimization algorithm. The establishment of the CNN-LSTM model optimized by the improved whale optimization algorithm involves connecting the output layer of the CNN model to the input gate of the LSTM model optimized by the improved whale optimization algorithm to obtain the WOA'-CNN-LSTM model.

[0034] Furthermore, the data processed in step D is substituted into the aforementioned model for training, and then risk level classification prediction is performed. This involves dividing the data preprocessed in step B and / or historical data into training and testing sets. The training set is substituted into the WOA'-CNN-LSTM model for training, and then the testing set is substituted into the trained WOA'-CNN-LSTM model for testing and correction. Finally, the preprocessed data is substituted into the corrected WOA'-CNN-LSTM model for instability risk level classification prediction.

[0035] In step E, accuracy Acc for: Specificity Spe for: Recall rate Rec for: F1 -Score for: In the formula: TP For a real example; FP This is a false positive example. TN A true counterexample; FN This is a false counterexample.

[0036] The mine goaf instability risk level classification and prediction system of the present invention includes a data acquisition module, a preprocessing module, a hyperparameter optimization module, an intelligent classification module, a classification evaluation module, and a result output module, the specific contents of which are as follows: The data acquisition module is used to acquire data related to the stability rating of goaf areas; The preprocessing module is used to preprocess the aforementioned data, including text digitization processing; The hyperparameter optimization module is used to optimize the hyperparameters of CNN-LSTM using an improved whale optimization algorithm. The intelligent classification module is used to build an improved CNN-LSTM model based on the improved whale optimization algorithm. The processed data is fed into the aforementioned model for training and risk level classification prediction. The classification evaluation module is used to assess accuracy. Acc Specificity Spe Recall rate Rec F1 -Score The aforementioned models are evaluated, and the superior model is selected based on the evaluation results. Then, the predicted risk level classification result of the superior model is used as the final result. The results output module is used to process the final results into text and obtain intelligent classification results.

[0037] The specific steps for improving the whale optimization algorithm in the hyperparameter optimization module are as follows: C10. Introduce Logistic chaotic mapping to initialize the population: In the formula: i =1,2,…,N are the population individual indices; j =1,2,…,D are dimension indices; μ These are the parameters for controlling chaos. K This represents the number of preheating iterations. u Let be a random variable uniformly distributed in the interval [0,1]. For the first i Initial values ​​for each individual; for i Individuals pass through k The chaotic value after the next chaotic iteration; for i Individuals pass through k The chaotic value after -1 chaotic iterations; lb j For the first j The lower bound of a dimension; ubj For the first j The upper bound of a dimension; Indicates the first i The individual in the first j The position of the dimension; After preheating j The chaotic value corresponding to the dimension; C20. Add the Levy flight strategy to the prey search phase of the whale optimization algorithm, as shown in the following formula: C21. Random Vector Generation: C22. Calculation of scale parameters and step size: C23, Location Update: In the formula: Represents random variables u Follows a mean of 0 and a variance of The normal distribution For scale parameters; Represents random variables v Follows a mean of 0 and a variance of The normal distribution Γ is the Gamma function; Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; Indicates the location of a random individual; A and C For coefficient vectors; Indicates the first t During the nth iteration i The individual in the first j The position of the dimension; α This is an adaptive step size factor; step Let levy be the step size; ub , lb Define the upper and lower boundaries of the algorithm; C30. A nonlinear adjustment development strategy is added to the spiral contraction stage of the whale optimization algorithm, as shown in the following formula: In the formula: Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; The optimal position is known. u 1. u 2. u 3 is a random number uniformly distributed within the interval [0,1].

[0038] The present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for classifying and predicting the risk level of instability in goaf areas of mines.

[0039] Example

[0040] like Figures 1 to 5 As shown, the specific steps for classifying and predicting the instability risk level of an existing goaf in a mine in Yunnan are as follows: S100: Obtain relevant data on existing goaf areas in the mine, including: burial depth, exposed roof area, goaf volume, goaf dip angle, adjacent goaf conditions (no other goaf areas within the affected area, goaf area is average and few in number, goaf area is large and numerous and scattered, goaf area is large and numerous and concentrated), rock compressive strength, geological structure (no faults, folds with little fold influence, faults partially cutting or folds having a large influence, faults penetrating the surrounding rock), rock mass structure (complete blocky structure, layered structure, fractured structure, loose structure), and goaf instability risk level (very low instability risk, low instability risk, high instability risk, very high instability risk), etc., which are 9 data related to the stability of the goaf.

[0041] S200: First, perform text digitization processing on the aforementioned text data such as geological structure, rock mass structure, and instability risk level. The geological structures without faults, with folds but minimal fold influence, partially cut by faults or significantly affected by folds, and with faults penetrating the surrounding rock are digitized as 1, 2, 3, and 4, respectively.

[0042] The complete blocky structure, layered structure, fractured structure, and loose structure in the rock mass are digitized as 1, 2, 3, and 4, respectively.

[0043] The instability risk levels are categorized as very small, low, relatively large, and very large, and are correspondingly numbered 1, 2, 3, and 4.

[0044] The following are the digitizations for the adjacent goaf situation: no other goaf within the affected area, goaf area of ​​average size and small number, goaf area of ​​large size and large number and scattered distribution, and goaf area of ​​large size and large number and concentrated distribution, corresponding to 1, 2, 3, and 4.

[0045] Then, the relevant data of existing goaf areas in the mine and the digitized text were processed to remove missing values ​​and outliers, and then the data was normalized to obtain 105 goaf stability-related data (see Table 1).

[0046]

[0047] S300, such as Figure 2As shown, the Whale Optimization Algorithm (WOA) is first improved, and the specific process is as follows: S310. Initialize the population by introducing a Logistic chaotic mapping: In the formula: i =1,2,…,N are the population individual indices; j =1,2,…,D are dimension indices; μ These are the parameters for controlling chaos. K This represents the number of preheating iterations. u Let be a random variable uniformly distributed in the interval [0,1]. For the first i Initial values ​​for each individual; for i Individuals pass through k The chaotic value after the next chaotic iteration; for i Individuals pass through k The chaotic value after -1 chaotic iterations; lb j For the first j The lower bound of a dimension; ub j For the first j The upper bound of a dimension; Indicates the first i The individual in the first j The position of the dimension; After preheating j The chaotic value corresponding to the dimension.

[0048] S320. In the prey search phase of the whale optimization algorithm, a Levy flight strategy is added, as shown in the following formula: S321, Random Vector Generation: S322, Calculation of scale parameters and step size: S323, Location Update: In the formula: Represents random variables u Follows a mean of 0 and a variance of The normal distribution For scale parameters; Represents random variables v Follows a mean of 0 and a variance of The normal distribution Γ is the Gamma function; Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; Indicates the location of a random individual; A and C For coefficient vectors; Indicates the first t During the nth iteration i The individual in the first j The position of the dimension; α This is an adaptive step size factor; step Let levy be the step size; ub , lb This represents the upper and lower boundaries of the algorithm.

[0049] S330. A nonlinear adjustment development strategy is added to the spiral contraction stage of the whale optimization algorithm, as shown in the following formula: In the formula: Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; The optimal position is known. u 1. u 2. u 3 is a random number uniformly distributed within the interval [0,1].

[0050] Then, the improved Whale Optimization Algorithm (WOA') was used to optimize the number of hidden node layers, initial learning rate, and L2 regularization coefficient of the LSTM model.

[0051] S400. Connect the output layer of the CNN model to the input gate of the aforementioned WOA'-optimized LSTM model to obtain the WOA'-CNN-LSTM model. Then, divide the preprocessed data and / or historical data into training and testing sets. Then, substitute the training set into the WOA'-CNN-LSTM model for training, and then substitute the testing set into the trained WOA'-CNN-LSTM model for testing and correction. Finally, substitute the preprocessed data into the corrected WOA'-CNN-LSTM model to perform instability risk level classification prediction.

[0052] S500: Simultaneously employs the WOA-CNN-LSTM model to predict the preprocessed data, and uses the accuracy... Acc Specificity Spe Recall rate Rec F1 -ScoreThe WOA-CNN-LSTM and WOA'-CNN-LSTM models were evaluated using different metrics. The better model was selected based on the evaluation metrics (the model with an evaluation metric close to 1 is the optimal one). The preprocessed data was then substituted into the better model to predict the instability risk level, and the predicted instability risk level classification results were used as the final results.

[0053] Among them: accuracy Acc for: Specificity Spe for: Recall rate Rec for:,

[0054] F1 -Score for:, In the formula: TP For a real example; FP This is a false positive example. TN A true counterexample; FN This is a false counterexample.

[0055] Accuracy of the two models Acc Specificity Spe Recall rate Rec F1 -Score The calculation results of the indicators are shown in Table 2. For example, Figure 3 As shown.

[0056] Table 2 Risk Level Classification Prediction and Assessment Indicators

[0057] As shown in Table 2, the WOA'-CNN-LSTM model has better risk level classification prediction evaluation index. Therefore, the WOA'-CNN-LSTM model is selected as the better model. Then, the preprocessed data is substituted into the WOA'-CNN-LSTM model to predict the instability risk level, and the instability risk level classification result is taken as the final result.

[0058] S600: The risk level classification prediction results of the aforementioned WOA-CNN-LSTM model and WOA'-CNN-LSTM model are respectively processed into text (that is, the digitized prediction results are processed into text), to obtain the prediction results of the WOA-CNN-LSTM model as follows: Figure 4 As shown, the prediction results of the WOA'-CNN-LSTM model are as follows: Figure 5 As shown.

[0059] Depend on Figure 4 and Figure 5It can be seen that the WOA'-CNN-LSTM model has a higher accuracy in classifying and predicting the stability of goaf areas.

[0060] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for classifying and predicting the instability risk level of goaf areas in mines, characterized in that: The process includes data acquisition, preprocessing, hyperparameter optimization, intelligent classification, classification evaluation, and result output. The specific steps are as follows: A. Data Acquisition: Acquire relevant data for stability rating of goaf areas; B. Preprocessing: Perform preprocessing on the aforementioned data, including text digitization; C. Hyperparameter optimization: The improved whale optimization algorithm is used to optimize the hyperparameters of CNN-LSTM; D. Intelligent classification: Build an improved CNN-LSTM model based on the improved whale optimization algorithm, input the processed data into the aforementioned model for training, and then perform risk level classification prediction; E. Classification Assessment: Using accuracy rate Acc Specificity Spe Recall rate Rec F1 -Score The indicators are used to evaluate the aforementioned models, and the better model is selected based on the evaluation indicators. Then, the predicted risk level classification result of the better model is used as the final result. F. Output Results: The final results are processed into text to obtain the intelligent classification results.

2. The method for classifying and predicting the risk level of instability in goaf areas according to claim 1, characterized in that: The data related to the stability rating of the goaf in step A includes burial depth, exposed roof area, goaf volume, goaf dip angle, adjacent goaf conditions, rock compressive strength, geological structure, rock mass structure, and instability risk level. In step B, the aforementioned data is preprocessed by first digitizing the text, then handling missing values ​​and outliers in the goaf stability rating data and the digitized text, and finally performing data normalization.

3. The method for classifying and predicting the risk level of instability in goaf areas according to claim 2, characterized in that: The geological structures, rock mass structures, and instability risk levels mentioned are all text data. The geological structures include no faults, folds with minimal fold impact, faults partially cutting through or folds significantly impacting the surrounding rock, and faults penetrating the surrounding rock. The rock mass structures include intact blocky structures, layered structures, fractured structures, and loose structures. The instability risk levels include very low instability risk, low instability risk, relatively high instability risk, and very high instability risk. The adjacent goaf conditions include no other goafs within the affected area, goafs with a moderate area and few in number, goafs with a large area and many in number and scattered distribution, and goafs with a large area and many in number and concentrated distribution. The text digitization process includes: The geological structures without faults, with folds but minimal fold influence, partially cut by faults or significantly affected by folds, and with faults penetrating the surrounding rock are digitized as 1, 2, 3, and 4, respectively. The complete blocky structure, layered structure, fractured structure, and loose structure in the rock mass are digitized as 1, 2, 3, and 4, respectively. The instability risk levels are categorized as very small, low, large, and very large, and assigned numerical values ​​of 1, 2, 3, and 4 respectively. The following are the digitizations for the adjacent goaf areas: no other goaf areas within the affected area, goaf areas of average size and few in number, goaf areas of large size and many in number and scattered, and goaf areas of large size and many in number and concentrated in number.

4. The method for classifying and predicting the risk level of instability in goaf areas according to claim 1, 2, or 3, characterized in that: The specific steps of the improved whale optimization algorithm are as follows: C10. Introduce Logistic chaotic mapping to initialize the population: In the formula: i =1,2,…,N are the population individual indices; j =1,2,…,D are dimension indices; μ These are the parameters for controlling chaos. K This represents the number of preheating iterations. u Let be a random variable uniformly distributed in the interval [0,1]. For the first i Initial values ​​for each individual; for i Individuals pass through k The chaotic value after the next chaotic iteration; for i Individuals pass through k The chaotic value after -1 chaotic iterations; lb j For the first j The lower bound of a dimension; ub j For the first j The upper bound of a dimension; Indicates the first i The individual in the first j The position of the dimension; After preheating j The chaotic value corresponding to the dimension; C20. Add the Levy flight strategy to the prey search phase of the whale optimization algorithm, as shown in the following formula: C21. Random Vector Generation: C22. Calculation of scale parameters and step size: C23, Location Update: In the formula: Represents random variables u Follows a mean of 0 and a variance of The normal distribution For scale parameters; Represents random variables v Follows a mean of 0 and a variance of The normal distribution Γ is the Gamma function; Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; Indicates the location of a random individual; A and C For coefficient vectors; Indicates the first t During the nth iteration i The individual in the first j The position of the dimension; α This is an adaptive step size factor; step Let levy be the step size; ub , lb Define the upper and lower boundaries of the algorithm; C30. A nonlinear adjustment development strategy is added to the spiral contraction stage of the whale optimization algorithm, as shown in the following formula: ， In the formula: Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; The optimal position is known. u 1. u 2. u 3 is a random number uniformly distributed within the interval [0,1].

5. The method for classifying and predicting the risk level of instability in goaf areas according to claim 4, characterized in that: The aforementioned optimization of CNN-LSTM using the improved whale optimization algorithm involves optimizing the number of hidden node layers, initial learning rate, and L2 regularization coefficient of the LSTM model using the improved whale optimization algorithm. The aforementioned method for establishing an improved CNN-LSTM model optimized by the whale optimization algorithm involves connecting the output layer of the CNN model to the input gate of the LSTM model optimized by the improved whale optimization algorithm, resulting in the WOA'-CNN-LSTM model.

6. The method for classifying and predicting the risk level of instability in goaf areas according to claim 5, characterized in that: The data processed in step D is substituted into the aforementioned model for training, and then risk level classification prediction is performed. This involves dividing the data and / or historical data preprocessed in step B into training and testing sets. The training set is substituted into the WOA'-CNN-LSTM model for training, and then the testing set is substituted into the trained WOA'-CNN-LSTM model for testing and correction. Finally, the preprocessed data is substituted into the corrected WOA'-CNN-LSTM model for instability risk level classification prediction.

7. The method for classifying and predicting the risk level of instability in goaf areas according to claim 4, characterized in that: In step E, accuracy Acc for: Specificity Spe for: Recall rate Rec for: F1 -Score for: In the formula: TP For a real example; FP This is a false positive example. TN A true counterexample; FN This is a false counterexample.

8. A classification and prediction system for the instability risk level of a mine goaf, characterized in that: The process includes data acquisition, preprocessing, hyperparameter optimization, intelligent classification, classification evaluation, and result output modules, detailed as follows: The data acquisition module is used to acquire data related to the stability rating of goaf areas; The preprocessing module is used to preprocess the aforementioned data, including text digitization processing; The hyperparameter optimization module is used to optimize the hyperparameters of CNN-LSTM using an improved whale optimization algorithm. The intelligent classification module is used to build an improved CNN-LSTM model based on the improved whale optimization algorithm. The processed data is fed into the aforementioned model for training and risk level classification prediction. The classification evaluation module is used to assess accuracy. Acc Specificity Spe Recall rate Rec F1 -Score The aforementioned models are evaluated, and the superior model is selected based on the evaluation results. Then, the predicted risk level classification result of the superior model is used as the final result. The results output module is used to process the final results into text and obtain intelligent classification results.

9. The mine goaf instability risk level classification and prediction system according to claim 8, characterized in that: The specific steps for improving the whale optimization algorithm in the hyperparameter optimization module are as follows: C10. Introduce Logistic chaotic mapping to initialize the population: In the formula: i =1,2,…,N are the population individual indices; j =1,2,…,D are dimension indices; μ These are the parameters for controlling chaos. K This represents the number of preheating iterations. u Let be a random variable uniformly distributed in the interval [0,1]. For the first i Initial values ​​for each individual; for i Individuals pass through k The chaotic value after the next chaotic iteration; for i Individuals pass through k The chaotic value after -1 chaotic iterations; lb j For the first j The lower bound of a dimension; ub j For the first j The upper bound of a dimension; Indicates the first i The individual in the first j The position of the dimension; After preheating j The chaotic value corresponding to the dimension; C20. Add the Levy flight strategy to the prey search phase of the whale optimization algorithm, as shown in the following formula: C21. Random Vector Generation: C22. Calculation of scale parameters and step size: C23, Location Update: In the formula: Represents random variables u Follows a mean of 0 and a variance of The normal distribution For scale parameters; Represents random variables v Follows a mean of 0 and a variance of The normal distribution Γ is the Gamma function; Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; Indicates the location of a random individual; A and C For coefficient vectors; Indicates the first t During the nth iteration i The individual in the first j The position of the dimension; α This is an adaptive step size factor; step Let levy be the step size; ub , lb Define the upper and lower boundaries of the algorithm; C30. A nonlinear adjustment development strategy is added to the spiral contraction stage of the whale optimization algorithm, as shown in the following formula: In the formula: Indicates the first t At the +1st iteration, the... i The individual in the first j The position of the dimension; The optimal position is known. u 1. u 2. u 3 is a random number uniformly distributed within the interval [0,1].

10. A computer-readable storage medium, characterized in that: It stores a computer program, which, when executed by a processor, implements the method for classifying and predicting the risk level of instability in goaf areas as described in any one of claims 1 to 7.

Images