Method for evaluating deformation of goaf collapse tower based on RF-SVM (Radio Frequency-Support Vector Machine)

Abstract

The invention discloses a goaf collapse tower deformation evaluation method based on an RF-SVM. The method comprises the steps that 1, the surface gradient, the coal seam inclination angle, the mining depth and thickness ratio, the goaf length, the mining layer number, the coal seam spacing and the horizontal distance between a tower and a maximum settlement point are selected to form an original evaluation factor system; step 2, performing normalization and standardization preprocessing on each piece of data; 3, determining the weight of each feature factor; 4, the screened evaluation factors are used for training an SVM model; step 5, performing prediction by using the trained SVM model; the problem that the disasters are difficult to accurately and timely discover and prevent by the traditional monitoring means is solved.

Description

Technical Field

[0001] The invention belongs to the technical field of transmission tower monitoring, and in particular relates to a method for evaluating the deformation of a mined-out collapsed pole tower based on RF-SVM. Background Art

[0002] As the most basic supporting project and infrastructure for industrial and mining enterprises in mining areas, high-voltage transmission lines cover almost all the tops of mining wells. Under the influence of ground subsidence in goaf areas, the foundations of these transmission lines are prone to settlement, horizontal displacement, tilt, and uneven settlement, which leads to the instability of the tower structure and seriously threatens the safe operation of the transmission lines. Guizhou Province is located in the Yunnan-Guizhou Plateau, with complex and diverse terrain. Goaf subsidence can easily induce geological disasters, including landslides, collapses, and mud-rock flows. These geological disasters occur frequently, posing a long-term threat to power facilities.

[0003] Since the power transmission lines of the power grid were put into operation, the power transmission and distribution facilities have been affected by geological disasters to varying degrees every year. Hidden dangers such as landslides, debris flows, collapses, ground fissures, ground collapses, and slope instability frequently occur, causing losses of up to tens of millions of yuan each year. Geological disasters in power transmission and distribution facilities are characterized by small size, high dispersion, weak deformation and damage characteristics, severe cloud and fog interference, and strong concealment. Traditional monitoring methods are difficult to accurately and timely detect and prevent these disasters. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method for evaluating the deformation of a mining tower based on RF-SVM to solve the technical problem that traditional monitoring means are difficult to accurately and timely discover and prevent these disasters.

[0005] The technical solution of the present invention is:

[0006] A method for evaluating the deformation of a mining tower based on RF-SVM, the method comprising:

[0007] Step 1: Select the surface slope, coal seam inclination, mining depth and thickness ratio, goaf length, mining layer number, coal seam spacing and horizontal distance between tower and maximum settlement point to form the original evaluation factor system;

[0008] Step 2: preprocessing the data by normalization and standardization;

[0009] Step 3: Determine the weight of each characteristic factor;

[0010] Step 4: Use the screened evaluation factors to train the SVM model;

[0011] Step 5: Use the trained SVM model for prediction.

[0012] The method for preprocessing the normalization and standardization of each data includes: the normalization and standardization calculation formula is as follows:

[0013] Standardization:

[0014] In the formula, x new represents the data after standardization; x i represents the data without standardization; μ represents the mean of the data; σ represents the standard deviation (std);

[0015] Normalization:

[0016] In the formula, x new represents the normalized data; x i represents the data without normalization; x max Indicates the maximum value of the same set of sample data; x min Indicates the minimum value of the same set of sample data.

[0017] Methods for determining the weights of each feature factor include:

[0018] Step 3.1: Use bootstrap sampling to extract N samples with replacement from the given original training set, and ensure that the number of features of each sample is consistent with that of the original training set;

[0019] Step 3.2: For N samples, respectively establish decision tree models and obtain N classification results;

[0020] Step 3.3: For each sample, randomly select m feature values ​​from the total feature n as the split feature set, and select the optimal feature to grow the node. When m < n, there are differences between the decision trees, and the calculation is finally completed;

[0021] Step 3.4: Use grid search hyperparameter optimization to obtain the optimal parameter combination;

[0022] Step 3.5, evaluate the advantages and disadvantages of the random forest model under different parameter combinations. The evaluation indicators include R-squared value, mean absolute error (MAE) and mean square error (MSE); R-squared value is an indicator of data fitting degree, and its distribution range is [0-1]. The closer it is to 1, the better the fitting degree; the mean absolute error (MAE) refers to the average difference between the true value and the fitted value. The closer the value is to 0, the more appropriate the analysis result of the model; the mean square error (MSE) represents the mean sum of squared errors between data. The smaller the value, the better the model.

[0023] The characteristic weights of the surface slope, coal seam inclination, mining depth and thickness ratio, goaf length, number of mining layers, coal seam spacing and horizontal distance between the tower and the maximum settlement point are 0.22, 0.13, 0.33, 0.1, 0.17, 0.04 and 0.013 respectively.

[0024] The method of using the screened evaluation factors for training the SVM model includes: firstly, dividing the data set into two parts, namely a training set and a test set, wherein the training set accounts for 70% of the total number of samples and the test set accounts for 30% of the total number of samples; inputting the 6 evaluation factors obtained after screening into the training SVM model, and using the scikit-learn library in python to complete the model establishment; in order to optimize the hyperparameters of the SVM model, grid search is used for hyperparameter optimization, by systematically searching for the best combination within the predefined parameter range, selecting the best performing parameter combination to train the model, and finally obtaining the best hyperparameters.

[0025] Beneficial effects of the present invention:

[0026] The present invention can provide early warning at the early stage of a disaster, helping supervisors to eliminate hidden dangers in a timely manner, thereby effectively ensuring the stability and safety of power facilities.

[0027] The present invention conducts specialized research on the characteristics and laws of surface deformation in goaf subsidence areas, analyzes the degree of surface deformation under different combinations of geological environment conditions and underground mining conditions, and judges the susceptibility of foundation deformation at the location of the tower based on the study of surface susceptibility, providing a scientific basis for further research on the evolution process of tower deformation hazard in goaf areas, which has urgent engineering application value and important practical significance.

[0028] It solves the problem that traditional monitoring methods are difficult to accurately and timely detect and prevent these disasters. DETAILED DESCRIPTION

[0029] Random Forest (RF) is a commonly used ensemble learning algorithm, which consists of multiple decision trees. It combines the modeling ability of decision trees and the advantages of ensemble learning, and is highly flexible and robust. In the process of building a decision tree, a hierarchical decision tree is constructed by recursively dividing the feature space into smaller subspaces. After the decision tree is built, each decision tree in RF can be used to predict new samples. In RF, the sampling process includes using the bootstrap sampling method to randomly extract samples from the original data set to form a subset, and when building each decision tree, some features are randomly selected for splitting. This randomness ensures the difference between each tree and enhances the generalization ability of the ensemble model.

[0030] When performing classification problems, RF determines the final classification result by voting. Each decision tree predicts the new sample and selects the category with the most occurrences as the final classification based on the voting results, thereby effectively reducing the risk of overfitting and showing good performance in various data sets and problems.

[0031] Support vector machine (SVM) constructs a hyperplane or a set of hyperplanes in high-dimensional or infinite-dimensional space, which can be used for classification, regression or other. In the field of regression, support vector regression (SVR) transforms the support vector machine to solve the classification problem into the regression fitting problem.

[0032] As an algorithm for finding segmentation hyperplanes in high-dimensional space, its theoretical basis is relatively complete. In the face of geological disasters with complex formation mechanisms, it can perform complex nonlinear classification through kernel functions. In the evaluation of geological disaster susceptibility, the particularity of the disaster determines that its sample data is limited, it cannot be intervened by humans, and it has a nonlinear relationship with each evaluation factor. SVM can better adapt to these problems. Based on the above core ideas, the introduction of SVM can efficiently solve the problem of nonlinear classification of high-dimensional data under the premise of a limited number of learning samples.

[0033] The present invention selects engineering geological factors, mining factors and factors of the position relationship between towers and goafs to construct an influencing factor system based on the characteristics of power transmission lines and the occurrence conditions of coal mines in mountainous areas.

[0034] The preliminary selection method of disaster-causing factors: the engineering geological factors specifically select two factors, namely the surface slope and the coal seam inclination; the mining factors select four factors, namely the ratio of mining depth to mining thickness, the length of the goaf, the number of mining layers, and the spacing between coal seams; the tower-related factors select one factor, namely the relative position of the tower and the goaf (the horizontal distance between the tower and the maximum settlement point). A total of seven factors constitute the original evaluation factor system.

[0035] The present invention selects the above 7 factors to form the evaluation system because:

[0036] Surface slope: According to the displacement after underground coal seam mining, it can be seen that with the increase of surface slope, the impact range of mining expands, and the maximum displacement generated also expands. According to the displacement monitoring curve of the slope, when the surface slope is less than 30°, the slope displacement is inversely proportional to the surface slope. At this time, the change degree of the surface of the goaf is small. Since the surface slope is gentle at this time, a large area of ​​the slope will pass above the goaf, so the total displacement of the surface shows a trend of first increasing and then decreasing with the increase of the distance from the bottom of the slope; after the slope exceeds 30°, the relationship between the slope displacement and the surface slope becomes positively correlated. As the slope continues to increase, the stability of the surrounding rock continues to decrease, resulting in a significant increase in the total displacement of the slope surface. In summary, it is more feasible to use the surface slope as an influencing factor for the safety analysis of the transmission and distribution line tower. At the same time, it can be preliminarily suggested that when erecting the tower, the area with a surface slope below 45° should be selected as much as possible.

[0037] Coal seam inclination:

[0038] As the coal seam inclination increases, the impact range of coal seam mining gradually shifts to the direction perpendicular to the coal seam inclination, and the displacement of the ground surface on the left side of the goaf is also increasing. From the overall observation of displacement monitoring, it can be seen that the relationship between the total surface displacement on both sides of the goaf and the coal seam inclination is opposite. The total surface displacement on the left side of the goaf (i.e., uphill direction) is positively correlated with the coal seam inclination, while the total surface displacement on the right side of the goaf (i.e., downhill direction) is negatively correlated with the coal seam inclination. The displacement in the uphill direction changes relatively slowly, while the displacement in the downhill direction changes rapidly. When the coal seam inclination is less than 25°, the maximum surface displacement point is located near the center of the goaf. When the coal seam inclination exceeds 25°, the closer to the top of the slope, the greater the surface displacement and gradually decreases with the increase of distance. At this time, the maximum surface displacement will also increase with the increase of the coal seam inclination, and the maximum displacement point is no longer near the center of the goaf. It can be seen from this that the displacement and relative displacement between two adjacent towers will be different for different coal seam inclinations. Therefore, it is more feasible to use coal seam inclination as an influencing factor for the safety of transmission and distribution line towers.

[0039] Goaf length

[0040] In order to explore the feasibility of using goaf length as an influencing factor for transmission tower safety, four groups of models were constructed using Midas GTS software. The goaf lengths of each group of models were 100m, 200m, 300m, and 400m, respectively. Other factors were set as surface slope 8°, coal seam inclination 5°, and mining depth and thickness ratio 30. The mining conditions of underground coal seams under different goaf lengths were simulated. Starting from the top of the slope, a monitoring point was set every 50m on the slope to obtain the influence of goafs of different lengths on the total surface displacement.

[0041] According to the displacement monitoring curve and the displacement of the slope after mining, it can be seen that the length of the goaf has a significant impact on the surface displacement. The surface displacement shows a positive correlation with the length of the goaf. In addition, as the length of the goaf continues to increase, the increase in the surface displacement also increases. Especially after the length of the goaf exceeds 300m, the growth rate of the displacement increases sharply. In addition, since the overburden movement deformation and surface goaf collapse are transmitted from the bottom to the top of the goaf, as the length of the goaf continues to increase, the impact range of coal seam mining is also expanding. Therefore, the relative displacement of two adjacent points on the surface increases. This change is reflected in the transmission and distribution line towers, which may cause the sag offset between adjacent towers to increase. It can be determined that the length of the goaf will have a greater impact on the safety of the tower, so it is used as an influencing factor for the safety analysis of the transmission and distribution line towers.

[0042] Mining depth and thickness ratio:

[0043] According to the mining data collected, most of them adopt the long-arm mining method and use the full caving method to manage the coal seam roof. When the mining method and roof management method are the same, the main mining factor affecting the surface movement basin is the mining depth and thickness ratio. The mining depth refers to the depth of the coal mining face from the surface, and the mining thickness refers to the mining thickness of a single coal seam. Under the condition of multi-coal seam mining, if the distance between coal seams is less than the influence range of the lower coal seam, the multi-layer coal seams including the interlayers between them are regarded as one layer.

[0044] Most relevant literatures use 30 and 60 as the standard for grading the mining depth and thickness ratio, and divide the mining area into goaf and subsidence zones. Among them, when the mining depth and thickness ratio is less than 30, the surface deformation after coal mining is strong, and sudden discontinuous deformation phenomena such as step-shaped subsidence and large cracks will occur on the surface; when the mining depth and thickness ratio is in the range of 30 to 60, the surface deformation after mining is large, and the surface will generate continuous deformation, and the cracks are only distributed on the edge of the surface movement basin; when the mining depth and thickness ratio is greater than 60, the surface deformation caused by coal seam mining is relatively small. In addition, there are studies showing that under the same mining thickness, the surface movement and deformation gradually decrease as the mining depth continues to increase, but the scope of influence of surface deformation will increase, and the time required for the formation to reach stability after mining will also increase.

[0045] Midas GTS software was used to construct four models with mining depth and thickness ratios of 20, 30, 50 and 100 respectively. Other geological conditions and mining conditions in the models were the same. A monitoring point was set up on the slope every 50m from the top of the slope to record the changes in the displacement of the ground under the influence of different coal seam mining depth and thickness ratios, so as to judge the influence of mining depth and thickness ratio on the safety of transmission towers.

[0046] The simulation results show that the surface displacement caused by underground mining is significantly negatively correlated with the mining depth and thickness ratio. The relative displacement of two adjacent points decreases with the increase of the mining depth and thickness ratio. That is, the larger the mining depth and thickness ratio, the larger the range of overburden movement, but the deformation of the surface is relatively mild. On the contrary, the smaller the mining depth and thickness ratio, the smaller the range of overburden movement, but the deformation of the surface is aggravated, with a gathering phenomenon. Therefore, when analyzing the safety of power transmission and distribution line towers in mining-out subsidence areas, the mining depth and thickness ratio is an influencing factor that needs to be considered.

[0047] Relative position between tower and goaf:

[0048] The movement and deformation of the cross-section of the surface moving basin formed after the critical full mining of the horizontal coal seam are basically symmetrically distributed about the center point of the goaf. At this time, the maximum surface subsidence point after underground mining is located at the center of the surface moving basin; the inclination value gradually increases between the boundary of the moving basin to the inflection point, decreases between the inflection point and the maximum subsidence point, and decreases to 0 at the maximum subsidence point. The maximum inclination value is located at the inflection point; the horizontal movement has the same change law as the inclination value.

[0049] The surface of mountainous areas is undulating and the terrain is changeable. Therefore, the surface movement basin formed by mining is different from the surface movement basin in flat land. Specifically, the surface movement basin in mountainous areas is irregular in shape, has a larger impact range, and the settlement at the same point is also greater. According to current research, the horizontal movement curve of the surface in mountainous areas is a curve with a central maximum value biased downhill. In addition, from the surface displacement monitoring curves of the five factors mentioned above, it can be seen that when the surface is tilted, the direction of the maximum displacement will shift uphill, and the surface displacement on both sides of the center line of the goaf is no longer symmetrical, but instead shows a trend of being relatively gentle on the left side of the center line and relatively steep on the right side of the center line. Therefore, the position relationship of a certain point on the surface of the mountainous area relative to the goaf will have a great impact on the displacement of the point. Considering that the tower is extremely sensitive to the surface tilt deformation due to its own high height, the relative position of the tower and the goaf cannot be ignored when conducting a safety analysis of the transmission line.

[0050] The above factors are orthogonally combined to form a calculation model sample. It is assumed that three towers are set up on the slope. Through computer numerical simulation with Midas GTS software, the surface deformation value during the mining process of the goaf is calculated under different combinations of environmental geological conditions, mining characteristics, and assumed tower characteristics (different working conditions).

[0051] The numerical simulation results were integrated, and the maximum tilt and maximum sag change (i.e., the maximum relative sinking between towers) of the three towers in the working condition were used as dependent variables to form two sets of data sets. In order to adjust the scale of the data so that it can better fit the model, eliminate the inaccuracy caused by the small amount of data itself, and improve the performance of machine learning, the data were normalized and standardized.

[0052] The main data preprocessing methods are standardization and normalization. Standardization is the process of subtracting the mean of sample data and dividing it by the standard deviation. The processed data distribution has the characteristics of zero mean and unit variance. The standardized data distribution has zero mean and unit variance, and the model analysis converges faster, avoiding the adverse effects of scale differences of different features on the model. Normalization is to scale the sample data to a specific range, usually [0,1] or [-1,1]. The range of the data after normalization is limited to avoid instability in model training caused by differences in numerical ranges, thereby accelerating model training. The combined use of the two methods can better accelerate the training model and integrate the analysis of sample data. The calculation formulas for standardization and normalization are as follows:

[0053] Standardization:

[0054] In the formula, x new represents the data after standardization; x i represents the data without standardization; μ represents the mean of the data; σ represents the standard deviation (std).

[0055] Normalization:

[0056] In the formula, x new represents the normalized data; x i represents the data without normalization; x max Indicates the maximum value of the same set of sample data; x min Indicates the minimum value of the same set of sample data.

[0057] Method for determining characteristic factor weights:

[0058] Python programming is used to train the acquired standard data using the random forest algorithm. The specific process is as follows:

[0059] (1) Using bootstrap sampling, N samples are extracted with replacement from a given original training set, ensuring that the number of features of each sample is consistent with that of the original training set;

[0060] (2) For N samples, a decision tree model is established and N classification results are obtained;

[0061] (3) For each sample, randomly select m (m≤n) feature values ​​from the total feature n as the split feature set, and select the optimal feature to grow the node. When m<n, there are differences between the decision trees. Finally, the calculation is completed.

[0062] There are several parameters in the random forest function that have a great impact on the performance of the algorithm. They are the minimum number of samples for leaf nodes (min_samples_leaf), the minimum number of samples required for internal node division (min_samples_split), the maximum depth (max_depth), and the maximum number of features (max_features). Here, grid search hyperparameter optimization is used to obtain the optimal parameter combination. When min_samples_leaf is 2, min_samples_split is 1, and the number of decision trees is 300, the random forest model is optimal.

[0063] The evaluation of the quality of the random forest model under different parameter combinations mainly refers to the following three evaluation indicators: R-squared value, mean absolute error (MAE), and mean square error (MSE). The R-squared value is an indicator of the degree of data fit, and its distribution range is [0-1]. The closer it is to 1, the better the degree of fit. At the same time, it should be noted that the difference between the R-squared values ​​of the training set and the test set should not be too large. If the fitting index of the training set is significantly higher than that of the test set, it means that the model has an overfitting problem. The mean absolute error (MAE) refers to the average difference between the true value and the fitted value. The closer this value is to 0, the more appropriate the analysis results of the model are. The mean square error (MSE) represents the mean sum of squared errors between data. The smaller this value is, the better the model is.

[0064] Table 1: Model evaluation results

[0065]

[0066] At this time, the R-squared values ​​of the training set and the test set are both close to 1, and the difference between the two is less than 0.05, which means that the fitting effect of the model is very good; the MAS and MSE values ​​are also close to 0, especially for the test set, the values ​​of the two indicators are basically 0. It can be inferred that it is feasible to analyze the data using the random forest model under this parameter combination.

[0067] The present invention involves a total of 7 evaluation factors in the calculation, namely, surface slope, coal seam inclination, mining depth ratio, goaf length, horizontal distance between the tower and the maximum settlement point, number of mining layers, and coal seam spacing. The feature weights are 0.22, 0.13, 0.33, 0.1, 0.17, 0.04, and 0.013, respectively. Since the characteristic coal seam spacing accounts for a small proportion, it can be eliminated. The first 6 features after RF algorithm sorting are selected as the optimal input features of the diagnostic model.

[0068] A support vector machine (SVM) model was constructed to predict the settlement of the surface of the mining subsidence area so as to evaluate the susceptibility. First, the data set was divided into two parts, namely the training set and the test set, where the training set accounted for 70% of the total samples and the test set accounted for 30% of the total samples. The six evaluation factors obtained after screening were input into the training SVM model, and the susceptibility evaluation model was established using the scikit-learn library in python. SVM includes a variety of kernel functions, which are used to map data to a high-dimensional space to achieve nonlinear prediction. Among them, the commonly used kernel functions are linear kernel function, polynomial kernel function, radial basis kernel function and sigmoid kernel function. The radial basis kernel function maps the data to a high-dimensional feature space, so that the regression model can effectively process data with complex nonlinear relationships. For data sets with obvious nonlinear relationships, a more accurate regression model can be established. The radial basis kernel function has only one parameter, the parameter tuning is relatively simple, and the model training and prediction speed is fast. This time, the radial basis kernel function is selected, and the program needs to set two important parameters: the penalty factor C and the kernel function parameter gamma. Generally speaking, when the C and gamma values ​​are too small, the SVM model is prone to underfitting, resulting in a decrease in its generalization ability; when the C and gamma values ​​are too large, the SVM has poor tolerance for errors and is prone to overfitting.

[0069] In order to optimize the hyperparameters of the SVM model, a grid search method is used for hyperparameter optimization. The best combination is systematically searched within a predefined parameter range. The grid search performs cross-validation on each parameter combination to evaluate the model performance and avoid overfitting. Finally, the best performing parameter combination is selected to retrain the model to improve its accuracy in evaluating the susceptibility of the tower. This process ensures the stability and generalization ability of the model when dealing with complex nonlinear data. Finally, the optimal hyperparameters are obtained.

[0070] The evaluation of the quality of the random forest model under different parameter combinations mainly refers to the following three evaluation indicators: R-square value, mean absolute error value (MAE), and mean square error (MSE). Among them, the R-square value is an indicator of the degree of data fitting, and its distribution range is [0-1]. The closer it is to 1, the better the degree of fitting. At the same time, it should be noted that the difference between the R-square values ​​of the training set and the test set should not be too large. If the fitting index of the training set is significantly higher than that of the test set, it means that the model has an overfitting problem. The mean absolute error value (MAE) refers to the average difference between the true value and the fitted value. The closer this value is to 0, the more appropriate the analysis results of the model are. The mean square error (MSE) represents the mean sum of squared errors between data. The smaller this value is, the better the model is. The accuracy indicators of the model after training are as follows:

[0071] Table 2: Model evaluation results

[0072]

[0073] The prediction results show that the model has high prediction accuracy and can meet engineering needs.

[0074] Specific implementation cases:

[0075] Jiangxi Coal Mine is located in the east of Zhijin County, Guizhou Province, about 24 kilometers away from the county seat, and is under the jurisdiction of Huaqi Town, Zhijin County, Guizhou Province. The geographical coordinates of the mining area are: 106°00′02″~106°01′13″, 26°40′18″~26°41′31″ north latitude. The Zhijin-Huaqi highway passes through the north-west of the mining area, about 1 km away from the mine. There is a mining highway intersecting with the Zhijin-Huaqi highway. There is a S307 provincial highway 7 km west of the mining area. The mine is about 27 kilometers away from the nearest highway, and the mining area has convenient transportation.

[0076] The survey area belongs to the low-medium mountain tectonic dissolution-erosion landform with complex terrain. The highest point is located at the top of Gaopo Mountain in the northwest corner of the mining area, with an altitude of 1553.6m. The lowest point is located at the bottom of Xiaohedong Falls Cave in the southeast of the mining area, with an altitude of 1365m. The maximum relative height difference is 188.6m.

[0077] The exposed strata in the mining area are from old to new, including Emeishan basalt (P 3 β), the first section of Longtan Formation (P 3 l 1 ), the second section of Longtan Formation (P 3 l 2 ), the third member of the Longtan Formation of the Upper Permian (P 3 l 3 ), Dalong Formation + Changxing Formation of the Upper Permian System (P 3 c+d), the first section of Yelang Group (T 1 y 1 ), the second member of the Yelang Formation of the Lower Triassic (T1 y 2 ), Quaternary (Q). The exposed strata in the surrounding areas of the mining area include the third section of the Yelang Formation (T 1 y 3 ), the first section of the Yongningzhen Formation of the Lower Triassic (T 1 yn 1 ), Quaternary system (Q).

[0078] The main deformation area of ​​Jiangxi coal mine is located in Liujia Village, Huaqi Town, Zhijin County. In recent years, villagers' houses have cracked and deformed, roads have deformed and damaged, poles and towers have tilted, and the mountains near the village have cracked to varying degrees. After field investigation, the characteristics are as follows:

[0079] Villagers' houses cracked and deformed: Many houses of villagers within 90 to 170 meters east of the No. 26 pole tower were severely cracked, threatening the safety of villagers' lives and property. The villagers in the area have been relocated and the deformed houses have been demolished. The breeding farm about 120 meters northwest of the No. 26 pole tower was severely deformed, with cracks on the ground and partial collapse of the farm wall.

[0080] Road deformation and damage: The roads around the No. 25 and No. 26 towers showed varying degrees of deformation and cracking, most of which were distributed horizontally along the roads.

[0081] Tower tilt: Tower No. 25 suffered severe tilt and deformation in 2021 and has been reinforced.

[0082] Mountain cracks: Cracks began to appear on the hill next to Tower No. 26 since July 2023. Most of the cracks are northwest-oriented, with a length of 0.5 to 70 m, a width of 0 to 0.2 m, and a visible depth of 0 to 0.5 m.

[0083] Using drone orthophoto data and combining it with on-site investigation, we statistically analyzed the cracks currently visible in the area and found out the relationship between crack development and goaf. The cracking and deformation of villagers' houses, deformation and damage of roads, tilting and deformation of pole towers, and the cracking of varying degrees in the mountains near the villages are all caused by the movement and deformation of the overburden in the goaf of Jiangxi coal mines.

[0084] Based on the field investigation and goaf data, the characteristic data of the two towers 25# and 26# in the goaf area were obtained, as shown in Table 3. The characteristic factors are surface slope, coal seam inclination, mining depth ratio, goaf length, horizontal distance between the tower and the maximum settlement point, and number of mining layers.

[0085] Table 3 Characteristic factor data

[0086]

[0087] The above data set is input into the trained SVM model for calculation and prediction. The prediction results show that the deformation values ​​of tower 25 and tower 26 are 0.065m and 0.053m respectively. Refer to the classification table of the development degree of mining collapse in the "Geological Hazard Risk Assessment Code". It is considered that the 25th tower has reached the "strong development" level, and the 26th tower has reached the "medium development" level.

[0088] Referring to the InSAR analysis results of the mining area from 2021 to 2024, it was found that the cumulative LOS deformation of the area where Tower 25 is located reached 0.069m. The cumulative LOS displacement of Tower 26 reached 0.058m. This is similar to the predicted results. It can be considered that the prediction results of this model are relatively reliable.

Claims

1. A method for evaluating the deformation of a mining tower based on RF-SVM, characterized by: The method comprises: Step 1: Select the surface slope, coal seam inclination, mining depth and thickness ratio, goaf length, mining layer number, coal seam spacing and horizontal distance between tower and maximum settlement point to form the original evaluation factor system; Step 2: preprocessing the data by normalization and standardization; Step 3: Determine the weight of each characteristic factor; Step 4: Use the screened evaluation factors to train the SVM model; Step 5: Use the trained SVM model for prediction.

2. The method for evaluating deformation of a mining tower based on RF-SVM according to claim 1, characterized in that: The method for preprocessing the normalization and standardization of each data includes: the normalization and standardization calculation formula is as follows: In the formula, x new represents the data after standardization; x i represents data without standardization; μ represents the mean of the data; σ represents the standard deviation (std); In the formula, x new represents the normalized data; x i represents the data without normalization; x max Indicates the maximum value of the same set of sample data; x min Indicates the minimum value of the same set of sample data.

3. The method for evaluating the deformation of a mining tower based on RF-SVM according to claim 1, characterized in that: Methods for determining the weights of each feature factor include: Step 3.1: Use bootstrap sampling to extract N samples with replacement from the given original training set, and ensure that the number of features of each sample is consistent with that of the original training set; Step 3.2: For N samples, respectively establish decision tree models and obtain N classification results; Step 3.3: For each sample, randomly select m feature values ​​from the total feature n as the split feature set, and select the optimal feature to grow the node. When m < n, there are differences between the decision trees, and the calculation is finally completed; Step 3.4: Use grid search hyperparameter optimization to obtain the optimal parameter combination; Step 3.5, evaluate the advantages and disadvantages of the random forest model under different parameter combinations. The evaluation indicators include R-squared value, mean absolute error (MAE) and mean square error (MSE); R-squared value is an indicator of data fitting degree, and its distribution range is [0-1]. The closer it is to 1, the better the fitting degree; the mean absolute error (MAE) refers to the average difference between the true value and the fitted value. The closer the value is to 0, the more appropriate the analysis result of the model; the mean square error (MSE) represents the mean sum of squared errors between data. The smaller the value, the better the model.

4. The method for evaluating the deformation of a mining tower based on RF-SVM according to claim 1, characterized in that: The characteristic weights of the surface slope, coal seam inclination, mining depth and thickness ratio, goaf length, number of mining layers, coal seam spacing and horizontal distance between the tower and the maximum settlement point are 0.22, 0.13, 0.33, 0.1, 0.17, 0.04 and 0.013 respectively.

5. The method for evaluating the deformation of a mining tower based on RF-SVM according to claim 1, characterized in that: The method of using the screened evaluation factors for training the SVM model includes: firstly, dividing the data set into two parts, namely a training set and a test set, wherein the training set accounts for 70% of the total number of samples and the test set accounts for 30% of the total number of samples; inputting the 6 evaluation factors obtained after screening into the training SVM model, and using the scikit-learn library in python to complete the model establishment; in order to optimize the hyperparameters of the SVM model, grid search is used for hyperparameter optimization, by systematically searching for the best combination within the predefined parameter range, selecting the best performing parameter combination to train the model, and finally obtaining the best hyperparameters.

Images