Mine pressure prediction method based on Adaboost ensemble learning algorithm
By applying Adaboost integrated learning algorithm and empirical modal decomposition method in the roof pressure prediction of coal mine working faces, the problem of insufficient accuracy and reliability of mineral pressure prediction in the existing technology is solved, and a higher accuracy and robust roof pressure prediction is achieved.
Patent Information
- Application Number
- CN202411773739.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-05
- Publication Date
- 2025-05-06
AI Technical Summary
In the prior art, in the prediction of roof pressure of coal mine working faces, there are limitations in prediction accuracy and reliability, making it difficult to deeply analyze and mine data in complex underground environments.
The ore pressure prediction method based on Adaboost integrated learning algorithm is adopted. By obtaining historical ore pressure data for preprocessing and feature extraction, the nonlinear signal is decomposed into multiple modal function sequences using the empirical modal decomposition method, and the model is trained using Adaboost integrated learning algorithm for each modal sequence, and finally the weighted accumulation is added to obtain the training output.
It significantly improves the prediction accuracy and reliability of the mine pressure prediction model, avoids the overfitting problem of a single prediction model, provides more robust and accurate roof pressure prediction results, and ensures safe mining of coal mine working faces.
Smart Images
Figure CN119939141A_ABST
Abstract
Description
Technical Field
[0001] The invention provides a mine pressure prediction method based on Adaboost ensemble learning algorithm, belonging to the technical field of coal mine roof pressure analysis and prediction. Background Art
[0002] At present, the coal industry is an important pillar industry of my country's economy and plays an important role in the country's economic development. However, in the process of coal mining, due to the harsh and complex underground environment, and the influence of factors such as complex ground stress, the roof pressure is constantly changing during the advancement of the working face, which brings major hidden dangers to coal mine safety. In serious cases, it can cause various disasters such as face spalling, frame compression, roof fall, rock burst, etc., which seriously affect the safe mining of coal. This not only threatens the safety of personnel and equipment, but also causes significant economic losses. The monitoring, analysis, management and prediction of the roof pressure of the coal mining working face have become an important part of coal mine safety production work.
[0003] In recent years, with the improvement of automation and informatization of coal mine working faces, the safety and production efficiency of working faces have been significantly improved. More and more working faces are equipped with electro-hydraulic controlled hydraulic supports. The change of working resistance of hydraulic supports reflects the mechanical characteristics of the overlying rock formations. Therefore, analyzing and predicting the roof pressure data collected by hydraulic supports is becoming an effective means to achieve advance warning and early response to working face pressure.
[0004] However, the mining environment of underground working faces changes in a complex manner. Traditional mine pressure prediction methods mainly rely on mechanical modeling and empirical formulas, which often require a large amount of field data support. The analysis of the data is only in the preliminary stage and has not reached the level of deeper analysis and mining of the data. The mine pressure prediction model established by ordinary machine learning methods (such as support vector machines) has certain limitations in prediction accuracy. Therefore, how to further improve the prediction accuracy and reliability of the mine pressure prediction model is a very meaningful research. Summary of the invention
[0005] In order to solve the above technical problems, the present invention proposes a mine pressure prediction method based on Adaboost ensemble learning algorithm.
[0006] The technical solution adopted by the present invention is: a mine pressure prediction method based on Adaboost ensemble learning algorithm, comprising the following steps:
[0007] Step 1: Obtain historical mine pressure data of the coal mine working face;
[0008] Step 2: preprocessing the collected historical mine pressure data;
[0009] Step 3, selecting a period of historical working cycle of mine pressure data to construct a training set;
[0010] Step 4: extract features from the training set, and use the empirical mode decomposition method to perform modal decomposition on the mine pressure data in the training set to obtain multiple modal function sequences;
[0011] Step 5: For each modal sequence function, select the corresponding ensemble learning algorithm and set various parameters of the ensemble learning algorithm;
[0012] Step 6: For each modal function sequence, the ensemble learning algorithm selected in step 5 is used to train the model;
[0013] Step 7: weightedly add and synthesize the training output results of each modal function sequence to obtain the training output of the training set;
[0014] Step 8, calculating the error statistics between the training output pressure data of the training set and the real pressure data;
[0015] Step nine, judging whether the trained model is the final model for mine pressure prediction according to the error statistics;
[0016] Step 10: construct a prediction sample set, input the prediction sample set into the final mine pressure prediction model, and obtain the prediction output value.
[0017] Furthermore, the mine pressure in step one is the roof pressure, which is collected by a pressure sensor installed in the lower cavity of the hydraulic support.
[0018] Furthermore, the preprocessing in step 2 includes testing the pure randomness and stationarity of the data, correcting the sampling interval of the data, processing missing values, removing redundancy from the data, and processing abnormal noise values.
[0019] Furthermore, the training set in step three includes multiple training data, each of which uses the mine pressure data at a historical moment as an input variable and the pressure data at the next moment in the future as an output variable.
[0020] Furthermore, in step 4, the empirical mode decomposition method is used to extract features of the input variables in the training set.
[0021] Furthermore, in steps 4 and 5, the ensemble learning algorithm for each modal function sequence is Adaboost, and the weak learner in Adaboost can select a variety of machine learning algorithms.
[0022] Furthermore, in step nine, if the error statistic is less than the preset threshold, the trained model is used as the final model for mine pressure prediction. Otherwise, return to step five, reset the various parameters of the integrated learning algorithm, and continue to generate a new mine pressure prediction model by training the training set.
[0023] Furthermore, if the number of returns is greater than a preset threshold, return to step three, reconstruct the training set, segment the training set, retrain only one segment of the training set or retrain several segments of the training set in sequence to obtain a segmented model.
[0024] Furthermore, the prediction sample set in step 10 is the mine pressure data of another historical working cycle in a different period from the training set, which is used to verify the accuracy of the final mine pressure prediction model.
[0025] Furthermore, the prediction sample set in step 10 is input into the final model of mine pressure prediction after feature extraction and empirical mode decomposition in step 4.
[0026] Compared with the prior art, the present invention has the following beneficial effects: the mine pressure prediction method based on the Adaboost ensemble learning algorithm of the present invention can further effectively improve the prediction accuracy of the common prediction algorithm model and has a wide range of applications. The specific advantages are as follows:
[0027] (1) The method of the present invention has certain universality and strong operability, and provides an effective basis for the prediction and early warning of the roof pressure of the working face. It has guiding significance for the selection of supports, the prevention and control of mine pressure disasters, and is beneficial to the safe production of coal mines.
[0028] (2) The method of the present invention does not directly send the collected mine pressure data to the learner for training, but instead extracts features from the mine pressure data and uses the empirical mode decomposition method to decompose the nonlinear and non-stationary mine pressure signal data into multiple single-frequency signal data. The patterns of these components changing over time are simpler, so they can be learned and trained separately, making it easier to reflect the regular changes of the roof pressure itself.
[0029] (3) The method of the present invention aims at the problem of overfitting caused by using a single prediction model in traditional machine learning methods. In order to further improve the prediction effect, the idea of ensemble learning is adopted. Adaboost is an ensemble learning method that can combine different weak learners of the same training set into a strong learner, so that the generalization ability of the weak learner model is improved, and the problem of overfitting of a single model is avoided, so as to obtain a more accurate and robust model. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] The present invention will be further described below in conjunction with the accompanying drawings:
[0031] Figure 1 is a flow chart of the method of the present invention;
[0032] Figure 2 It is a schematic diagram of the mine pressure data feature extraction and training model of the present invention;
[0033] Figure 3 It is a schematic diagram of the resistance change process of the hydraulic support of the coal mine working face;
[0034] Figure 4 Schematic diagram of the Adaboost ensemble learning model according to an embodiment of the present invention. DETAILED DESCRIPTION
[0035] like Figures 1 to 4 As shown, the present invention provides a mine pressure prediction method based on the Adaboost ensemble learning algorithm. According to the historical data of the pressure sensor of a hydraulic support collected from the coal mine working face, data preprocessing is performed, and a mine pressure data training set is constructed according to the data of a coal mining machine cutting coal in one cycle. The empirical mode decomposition method is used for feature extraction to obtain multiple modal function sequences. For each modal function sequence, a mine pressure prediction model based on Adaboost ensemble learning is established, and multiple training output results are obtained. The training output of the training set is obtained by weighted accumulation and synthesis. If the error statistic between the training output pressure data of the training set and the actual pressure data is less than the preset threshold, the training is completed, so that the roof pressure can be predicted, which specifically includes the following steps:
[0036] Step 1: Obtain historical mine pressure data of the coal mine working face, wherein the historical mine pressure data is essentially a time series signal;
[0037] Specifically, mine pressure refers to roof pressure. A pressure sensor is installed in the lower cavity of the hydraulic support in the coal mine. The pressure sensor is the column pressure sensor of the hydraulic support in the coal mine. It actually measures the pressure value of the liquid in the column of the hydraulic support, which can indirectly reflect and measure the pressure of the hydraulic support from the coal and rock on the upper roof. The pressure sensor can continuously monitor the pressure changes of the liquid in the lower cavity of the column and provide real-time data on the interaction between the hydraulic support and the coal mine roof.
[0038] Assume that the time series data collected for a hydraulic support is: 1 ,x 2 ,......,x M , where M is the total number of collected data.
[0039] Step 2: preprocessing the collected mine pressure data;
[0040] The specific preprocessing processes and steps include: testing the pure randomness and stationarity of the data, correcting the sampling interval of the data, processing missing values, removing redundancy and denoising, etc.
[0041] Step 3, selecting the mine pressure data of a historical working cycle to construct a mine pressure data training set;
[0042] like Figure 3As shown, the coal mining machine in the underground working face of the coal mine will form multiple coal cutting cycles, and the mine pressure data of one of the historical working cycles is selected to construct the mine pressure data training set. 1 The mine pressure data is an input variable, and the next moment t 2 The pressure data is an output variable. For example, N historical mine pressure data are taken as input variables, and the pressure data at the next moment is the output variable. The constructed mine pressure data training set is {(x 1 ,y 1 ),(x 2 ,y 2 ),......,(x N ,y N )}, as shown in Table 1 below.
[0043]
[0044] Table 1. Mine pressure data training set.
[0045] Step four, extract features from the mine pressure data training set, and use the empirical mode decomposition method to perform modal decomposition on the mine pressure signal data in the training set to obtain multiple modal function sequences; the empirical mode decomposition method can decompose nonlinear and non-stationary composite signals and obtain multiple single-frequency signals, and the number of single-frequency signals is related to the initial composite signal itself.
[0046] Specifically, the input variable (x 1 ,x 2 ,......,x N ) to extract features and obtain n modal function sequences (IMF 1 ,IMF 2 ,......,IMF n ), where the number of modal function sequences n is related to the input variable composite signal of the initial mine pressure data training set, and the final component residual function can be ignored or retained. 1 ,x 2 ,......,x N ) is decomposed into four modal function sequences (IMF 1 ,IMF 2 ,IMF 3 ,IMF 4 ), as shown in Table 2 below.
[0047]
[0048] Table 2 Empirical mode decomposition table.
[0049] Step 5: For each modal function sequence, use the Adaboost ensemble learning algorithm, select a learner as the base learner, and set various parameters of the ensemble learning algorithm.
[0050] n modal function sequences (IMF 1 ,IMF 2 ,......,IMF n ) is a matrix data with N rows and n columns. Each column of data, that is, each modal function sequence is matched with the training output variables of the N mine pressure data training sets and input into an Adaboost ensemble learner. For example, the i-th modal function sequence IMF i and (y 1 ,y 2 ,y 3 ,......,y N ) are matched and input into the i-th Adaboost ensemble learner for training, where i = 1, 2, 3, ..., n.
[0051] For example: select support vector machine as the base learner, set the number of base learners T, the kernel function and penalty parameters of the support vector machine algorithm. Among them, the kernel function is very important for the support vector machine. The type, form and parameter changes of the kernel function will change the transformation mapping from the original input space to the feature space, thereby affecting the properties of the feature space. There are dozens of kernel functions available at present. The most commonly used kernel function is the radial basis kernel function (RBF), and its calculation formula is as follows:
[0052]
[0053] Where: γ, σ 2 is a non-zero constant.
[0054] For different modal function sequences, the number T of base learners, the kernel function and penalty parameters of the support vector machine algorithm may be different.
[0055] Step 6: For each modal function sequence, the model is trained using the Adaboost ensemble learning algorithm;
[0056] Ordinary weak learners have limited learning capabilities. The Adaboost ensemble learning algorithm uses a weak learning model to perform weighted training on data, obtains new weight coefficients through the data error rate and updates the data weights, repeats this process to obtain multiple updated weak learning models, and then combines multiple weak learning models to obtain a strong learning model. The specific algorithm steps are as follows:
[0057] (1) Import sample data {(x j ,yj )|j=1,2,……,N},initialize the weight of each sample: D 1 =(w 11 ,w 12 ,...,w 1j ,...,w 1N ), w 1j =1 / N, j=1,2,3,......,N, N is the number of training samples in the Adaboost ensemble learner.
[0058] (2) Use the training data with weight distribution to learn on the tth base learner and obtain the base learner f t (x), t = 1, 2, 3, ..., T, T is the total number of base learners.
[0059] (3) Using the base learner f t (x) Predict the training samples and calculate the prediction error ε of the base learner on the training samples j =y j -f t (x j ), maximum error ε max =|y j -f t (x j )|, relative error ε tj =y j -f t (x j ) / ε max , j=1,2,3,......,N.
[0060] (4) Calculate the learning error rate Calculate the weight coefficient α of the tth base learner according to the learning error rate t =0.5*log(1-ε t ) / ε t .w tj is the weight of the j-th quantity at time t.
[0061] (5) According to the weight coefficient α of the tth base learner t Adjust the weight of the next training sample Dj, where the weight of the jth quantity at time t+1 is Normalization factor
[0062] (6) Repeat steps (2)-(5) and perform the next iteration until the number of iterations T is reached.
[0063] (7) A strong learning model is obtained by combining the final weights and the model f fin(x) is the strong learning prediction function.
[0064] Step 7: weightedly add and synthesize the training output results of each modal function sequence to obtain the training output of the training set;
[0065] Specifically, the training output of each modal function sequence is (y (1) ,y (2) ,...,y (i) ,...,y (n) ), where y (i) =(y (i) 1 ,y (i) 2 ,......,y (i) N ), set the weight coefficient B of each modal function sequence = (b 1 ,b 2 ,......,b n ), the training output results of each modal function sequence are calculated and weighted synthesis is performed to obtain the training output of the training set: in,
[0066]
[0067] Step 8, calculating the error statistics between the training output pressure data of the training set and the real pressure data;
[0068] The training output pressure data of the training set is The actual pressure data is y, and the difference between the two You can also calculate various statistical values of the difference between the two, for example, calculate the sample standard deviation between the training output pressure data and the true pressure data of the training set
[0069] Step 9: If the error statistic is less than the preset threshold, the Adaboost ensemble learning algorithm model is regarded as the final model for predicting the mine pressure of the hydraulic support. Otherwise, return to step 5, reset the various parameters of the Adaboost ensemble learning algorithm, and continue to generate a new mine pressure prediction model by training the training set. If the number of returns is greater than the preset threshold, return to step 3, reconstruct the training set, segment the training set, retrain only one segment of the training set, or retrain several segments of the training set in turn to obtain a segmented model.
[0070] For example, the preset sample standard deviation threshold is 0.1. When the sample standard deviation s between the training output pressure data and the real pressure data of the pressure data training set at N moments during the entire working cycle is less than 0.1, the model based on the Adaboost ensemble learning algorithm is regarded as the final mine pressure prediction model of the hydraulic support. Otherwise, return to step 5, reset various parameters of the Adaboost ensemble learning algorithm, and continue to generate a new mine pressure prediction model by training the training set.
[0071] If the preset return number threshold is 5, when the return number is greater than 5, the training set can be segmented, for example, into 3 segments, and only the first segment of training data is trained, or the three segments of data can be trained separately in turn to obtain a segmented model.
[0072] Step ten, using the final mine pressure prediction model to predict the collected prediction sample set.
[0073] The collected mine pressure prediction samples are used to construct a prediction sample set. The construction method is the same as that of the training set. Next, feature extraction is performed on the prediction sample set. The empirical mode decomposition method is used to perform modal decomposition on the mine pressure signal data in the test sample to obtain multiple modal function sequences, which are input into the trained Adaboost ensemble learning algorithm final model to obtain the predicted output value.
[0074] The present invention is suitable for predicting and modeling the roof pressure of coal mine working faces. Compared with the traditional machine learning method that uses a single prediction model and leads to overfitting, the Adaboost ensemble learning method can combine multiple weak learners into a strong learner, thereby improving its generalization ability. It has the advantages of high prediction accuracy and a more robust model, further improving the reliability of the prediction results, and can ensure the safe mining of the working face.
[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A mine pressure prediction method based on Adaboost ensemble learning algorithm, characterized by: The steps include: Step 1: Obtain historical mine pressure data of the coal mine working face; Step 2: preprocessing the collected historical mine pressure data; Step 3, selecting a period of historical working cycle of mine pressure data to construct a training set; Step 4: extract features from the training set, and use the empirical mode decomposition method to perform modal decomposition on the mine pressure data in the training set to obtain multiple modal function sequences; Step 5: For each modal sequence function, select the corresponding ensemble learning algorithm and set various parameters of the ensemble learning algorithm; Step 6: For each modal function sequence, the ensemble learning algorithm selected in step 5 is used to train the model; Step 7: weightedly add and synthesize the training output results of each modal function sequence to obtain the training output of the training set; Step 8, calculating the error statistics between the training output pressure data of the training set and the real pressure data; Step nine, judging whether the trained model is the final model for mine pressure prediction according to the error statistics; Step 10: construct a prediction sample set, input the prediction sample set into the final mine pressure prediction model, and obtain the prediction output value.
2. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 1, characterized in that: The mine pressure in step 1 is the roof pressure, which is collected by a pressure sensor installed in the lower cavity of the hydraulic support.
3. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 1, characterized in that: The preprocessing in step 2 includes testing the pure randomness and stationarity of the data, correcting the sampling interval of the data, processing missing values, removing redundancy from the data, and processing abnormal noise values.
4. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 1, characterized in that: The training set in step three includes multiple training data, each of which uses the mine pressure data at a historical moment as an input variable and the pressure data at the next moment in the future as an output variable.
5. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 4 is characterized in that: In step 4, the empirical mode decomposition method is used to extract features from the input variables in the training set.
6. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 1, characterized in that: In steps 4 and 5, the ensemble learning algorithm for each modal function sequence is Adaboost, and the weak learner in Adaboost can select a variety of machine learning algorithms.
7. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 1, characterized in that: In step nine, if the error statistic is less than the preset threshold, the trained model is used as the final mine pressure prediction model. Otherwise, return to step five, reset the various parameters of the ensemble learning algorithm, and continue to generate a new mine pressure prediction model by training the training set.
8. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 7 is characterized in that: If the number of returns is greater than the preset threshold, return to step three, reconstruct the training set, segment the training set, retrain only one segment of the training set or retrain several segments of the training set in sequence to obtain a segmented model.
9. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 1, characterized in that: The prediction sample set in step 10 is the mine pressure data of another historical working cycle in a different period from the training set, which is used to verify the accuracy of the final mine pressure prediction model.
10. The method for predicting mine pressure based on Adaboost ensemble learning algorithm according to claim 9, characterized in that: The prediction sample set in step 10 is input into the final mine pressure prediction model after feature extraction and empirical mode decomposition in step 4.