A coal mine tunneling face gas emission amount prediction method based on a KPCA-POA-LSTM model
By using the KPCA-POA-LSTM model and optimizing the long short-term memory network with kernel principal component analysis and peacock optimization algorithm, the problem of high accuracy and speed in predicting gas outburst in coal mine tunneling faces was solved, the prediction error was reduced, and the prediction capability of the gas monitoring system was improved.
Patent Information
- Application Number
- CN202211014081.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-23
- Publication Date
- 2025-12-09
- Estimated Expiration
- 2042-08-23
AI Technical Summary
Existing methods for predicting gas outbursts at coal mine tunneling faces cannot achieve high accuracy and speed. Traditional methods cannot effectively cope with the suddenness and complexity of gas accidents at coal mine tunneling faces, and there is insufficient research on multidimensional data matrices, resulting in large prediction errors.
A KPCA-POA-LSTM model was adopted, and gas data was processed by dimensionality reduction through kernel principal component analysis. The peacock optimization algorithm was used to optimize the forget gate and input gate of the long short-term memory network, and a multidimensional state matrix was constructed and predicted using the Adam solver.
It has achieved highly accurate prediction of gas emission in coal mine tunneling faces, reduced prediction errors, improved prediction speed, and provided key technical support for coal mine tunneling face gas monitoring systems.
Smart Images
Figure CN115470887B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application provides a coal mine tunneling face gas emission amount prediction method based on a KPCA-POA-LSTM model, and belongs to the technical field of coal mine tunneling face gas prediction. BACKGROUND
[0002] In the coal industry, the regional coal mine tunneling face gas system still needs to be further researched. In addition to rock deformation, stress concentration, stress concentration, stress superposition, a multi-level and multi-directional integrated coal mine tunneling face gas emission amount prediction model should also be researched. The coal mine tunneling face gas emission amount has the characteristics of fuzziness, uncertainty and randomness, and such characteristics make the traditional prediction method unable to accurately predict the coal mine tunneling face gas emission amount. Based on the suddenness and complexity of the coal mine tunneling face gas accident, how to accurately master the coal mine tunneling face gas emission law and accurately and quickly predict the coal mine tunneling face gas is the key to controlling the large-scale emission of the coal mine tunneling face gas in advance. Based on the storage conditions of the coal mine tunneling face gas, the coal mine tunneling face gas emission has certain chaotic characteristics, and the information of the coal and the coal mine tunneling face gas geological system becomes a nonlinear system in which the superposition principle no longer holds. The coal mine tunneling face gas system is analyzed from various factors such as original coal mine tunneling face gas content, buried depth, coal thickness, mining height, recovery rate, adjacent layer coal mine tunneling face gas content and mining intensity. Some domestic scholars have analyzed the essential causes of coal mine tunneling face gas accidents based on large coal mine tunneling face gas accidents, and the defect of other objective factors in the coal mine sudden event model, and have made a detailed analysis according to each link of the system. Through the analysis and simulation of examples, preventive measures are proposed, and a coal mine tunneling face gas emission amount prediction system is established. A real-time coal mine tunneling face gas monitoring system is constructed. A coal mine tunneling face gas emission amount prediction model is constructed. In the aspect of coal mine tunneling face gas emission amount prediction, many scholars have made certain contributions. There is little research on the multi-dimensional data matrix in the prediction of coal mine tunneling face gas emission amount, and the coal mine tunneling face gas single-layer feedforward neural network model is not trained. Therefore, the model is established from the perspective of the multi-dimensional state of the coal mine tunneling face gas emission amount and the single-layer feedforward neural network, and the model is predicted. SUMMARY
[0003] The technical problem to be solved by the application is that the application provides a coal mine tunneling face gas emission amount prediction method based on a KPCA-POA-LSTM model, which can more accurately predict coal mine tunneling face gas data.
[0004] The technical scheme of the present application is: a coal mine tunneling working face gas emission amount prediction method based on a KPCA-POA-LSTM model, the specific steps of the method are as follows:
[0005] Step 1, initialize the initial data by using kernel principal component analysis KPCA, set the radial basis kernel function parameter, and calculate the characteristic vector and characteristic value, and calculate the cumulative contribution rate;
[0006] Step 2, build a POA-LSTM model, use peacock optimization algorithm POA to optimize the forget gate, input gate, input node, output gate, intermediate output and state unit of long short-term memory network, and reduce the gradient disappearance problem;
[0007] Step 3, build a multi-dimensional state matrix and select an attractive strategy, use the optimized long short-term memory network to train the multi-dimensional state matrix, according to the coal mine tunneling working face gas original data, use Adam to solve, and predict the coal mine tunneling working face gas result.
[0008] Further, the specific steps of step 1 are:
[0009] Step 1.1, build kernel principal component analysis, set matrix X, sample point x i , and use each column of X to represent a sample X=[x1,x2…,x N ] Each sample point x i is a K-dimensional column vector, there are N samples in X, introduce a symmetric matrix K, and make its eigenvectors constitute the entire input space K ij = φ(x i ) T [φ(x j )], wherein φ(x) is a mapping, i and j represent the row and column of the matrix space respectively;
[0010] Step 1.2, introduce a kernel function, wherein x and y are low-dimensional vectors, σ1 and γ1 are constants, and the mapping to a high-dimensional vector is a radial basis function kernel Get the data norm and calculate the eigenvectors of the matrix.
[0011] Further, the specific steps of step 2 are:
[0012] Step 2.1, use the search and approach mechanism of the female peacock to optimize; the following mathematical model is used to describe the approach mechanism of the female peacock:
[0013]
[0014] Where r5 represents a random number uniformly distributed in [0,1]; X Ph represents the position vector of the female peacock, t is the current iteration number, tmax is the maximum iteration number; and θ 0 and θ 1 are set as 0.1 and 1 respectively; X Pc1 , X Pc2 , X Pc3 , X Pc4 , X Pc5 are five individuals with the highest fitness in the peacock population, according to formula (4), when θ < 1 / 3, the female peacock approaches the selected peacock, which represents the local utilization in the search process; when θ > 1 / 3, the female peacock tends to move to the symmetric position of the selected peacock, which represents the global exploration in the search process;
[0015] Step 2.2, the peacock juvenile adaptive search behavior is used for optimization; the position update of the finder in each iteration is described as follows: as shown below:
[0016]
[0017]
[0018] Γ (x) = (x-1)! (4)
[0019] Where r 6 and r 7 respectively represent two different one-dimensional random vectors uniformly distributed in [-1, 1] and [0, 1]; γ represents a constant, the value of which is set as 1.5, and! represents factorial; the specific behavior of each peacock juvenile is modeled as:
[0020]
[0021]
[0022] Where r 8 represents a random number uniformly distributed in [0, 1]; X SPc and X PcC respectively represent the position vectors of the selected peacock and the peacock juvenile; α and δ represent two coefficient factors, which dynamically change with the iteration number, and can be defined by the following formula:
[0023]
[0024]
[0025] The above formulas (7) and (8) show that at the beginning of iteration, α is greater than δ, and the peacock juvenile mainly performs random search; when iteration ends, δ is greater than α, and the peacock juvenile converges to the optimal five solutions, i.e. the positions of the five optimal peacocks; α 0 = 0.9, α 1 = 1.4, and δ 0 and δ 1 are equal to 0.1 and 1 respectively, and the output parameters are optimized by the peacock search behavior.
[0026] The beneficial effects of the present application are:
[0027] 1. The application utilizes nuclear principal component analysis to reduce the dimensionality of the gas outburst data of the coal mine tunneling face and initialize; then uses peacock optimization algorithm to optimize the nodes of long short-term memory network, so that the parameters are regenerated continuously; constructs a multi-dimensional state matrix, uses long short-term memory network to map the multi-dimensional state matrix and selects sigmoid as the activation function and Adam as the solver; then uses the optimized long short-term memory network to predict the coal mine tunneling face gas data;
[0028] 2. The application can accurately predict the gas emission of the coal mine tunneling face, and the prediction method has faster prediction speed and lower error rate, which provides an effective prediction method for the coal mine tunneling face gas monitoring system and provides key technical support for building a digital monitoring system for the coal mine tunneling face gas. BRIEF DESCRIPTION OF DRAWINGS
[0029] Figure 1 The prediction algorithm detailed flow block diagram of the application is exemplarily shown;
[0030] Figure 2 The prediction algorithm flow chart of the application is exemplarily shown;
[0031] Figure 3 The coal mine tunneling face gas content basic index cumulative contribution chart in the embodiment of the application is exemplarily shown;
[0032] Figure 4 The prediction result comparison of the LSTM, POA-LSTM and PSO-LSTM prediction models in the embodiment of the application is exemplarily shown;
[0033] Figure 5 The prediction result absolute error comparison of the LSTM, POA-LSTM and PSO-LSTM prediction models in the embodiment of the application is exemplarily shown. DETAILED DESCRIPTION
[0034] In order to make the purpose, technical scheme and advantages of the embodiments of the application clearer, the following will be combined with the drawings of the embodiments of the application to further describe the application. Figure 1 The technical scheme in the embodiments of the application is described clearly and completely, obviously, the described embodiments are part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application. It should be noted that, in the case of no conflict, the embodiments in the application and the features in the embodiments can be combined with each other at will.
[0035] Embodiment 1, the present application proposes a coal mine tunneling working face gas emission amount prediction method based on KPCA-POA-LSTM model, which is implemented according to the following steps.
[0036] A coal mine tunneling working face gas emission amount prediction method based on KPCA-POA-LSTM model, the specific steps of the method are as follows:
[0037] Step 1, use kernel principal component analysis KPCA to initialize the initial data, set the radial basis kernel function parameter, and calculate the characteristic vector and characteristic value, and calculate the cumulative contribution rate;
[0038] Step 2, build a POA-LSTM model, use peacock optimization algorithm POA to optimize the forget gate, input gate, input node, output gate, intermediate output and state unit of long short term memory network, and reduce the gradient disappearance problem;
[0039] Step 3, build a multi-dimensional state matrix and select an attractive strategy, use the optimized long short term memory network to train the multi-dimensional state matrix, according to the coal mine tunneling working face gas original data, use Adam to solve, and predict the coal mine tunneling working face gas result.
[0040] Further, the specific steps of step 1 are:
[0041] Step 1.1, build kernel principal component analysis, set matrix X, sample point x i , and use each column of X to represent a sample X = [x1, x2…, x N ] Each sample point x i is a K-dimensional column vector, there are N samples in X, introduce a symmetric matrix K, and make its eigenvectors constitute the entire input space K ij = φ(x i ) T [φ(x j )], where φ(x) is a mapping, i and j represent the row and column of the matrix space respectively;
[0042] Step 1.2, introduce the kernel function, where x and y are low-dimensional vectors, σ1 and γ1 are constants, and map them to high-dimensional vectors using the radial basis function kernel Get the data norm and calculate the eigenvectors of the matrix.
[0043] Further, the specific steps of step 2 are:
[0044] Step 2.1, use the search and approach mechanism of female peacock to optimize; the following mathematical model is used to describe the approach mechanism of female peacock:
[0045]
[0046] where r5 represents a random number uniformly distributed in [0, 1]; X Ph represents the position vector of the female peacock, t is the current iteration number, t max is the maximum iteration number; and θ0and θ1are set to 0.1 and 1, respectively; X Pc1 , X Pc2 , X Pc3 , X Pc4 , X Pc5 are the five individuals with the highest fitness in the peacock population. According to formula (4), when θ < 1 / 3, the female peacock approaches the selected peacock, which represents the local utilization in the search process; when θ > 1 / 3, the female peacock tends to move to the symmetric position of the selected peacock, indicating global exploration in the search process;
[0047] Step 2.2, the peacock chicks' adaptive search behavior is used for optimization; the position update of the finder in each iteration is described as follows:
[0048]
[0049]
[0050] Γ(x) = (x - 1)! (4)
[0051] where r6and r7represent two different one-dimensional random vectors uniformly distributed in [-1, 1] and [0, 1], respectively; γ represents a constant, which is set to 1.5, and! represents factorial; the specific behavior of each peacock chick is modeled as:
[0052]
[0053]
[0054] where r8represents a random number uniformly distributed in [0, 1]; X SPc and X PcC represent the position vectors of the selected peacock and the peacock chick, respectively; α and δ represent two coefficient factors that dynamically change with the iteration number and can be defined by the following formula:
[0055]
[0056]
[0057] The above formula (7), (8) shows that when iteration starts, alpha is greater than delta, and the peacock baby mainly performs random search; when iteration ends, delta is greater than alpha, and the peacock baby converges to the optimal five solutions, that is, the optimal five peacock positions; alpha0=0.9, alpha1=1.4, and delta0 and delta1 are equal to 0.1 and 1 respectively, and the output parameters are optimized by the peacock search behavior.
[0058] First, the accompanying drawings are combined Figure 1 , Figure 2 , the embodiment 1 of the present application is described.
[0059] Due to the ground stress and rheology of the coal mine tunneling face gas, rock deformation, stress concentration, stress superposition, vibration wave and the like make the coal mine tunneling process close to the high tectonic strain energy containing coal mine tunneling face gas rock stratum and coal seam, and the special effect causes the coal mine tunneling face gas to overflow, so that the coal mine tunneling face gas continuously gushes out; and the rheological behavior of the coal mine tunneling face gas in space is complex, and in the relaxation, stress concentration and original stress three regions, a complex nonlinear system is often formed.
[0060] Select the coal mine tunneling face gas data as shown in Table 1, and the main influencing factors are: coal mine tunneling face gas pressure, desorption coefficient, relative coal mine tunneling face gas emission, coal mine tunneling face gas trend peak ratio, coal mine tunneling face gas trend peak average ratio, ground stress parameter and ground stress index.
[0061] Table 1 coal mine tunneling face gas data
[0062]
[0063]
[0064] The nuclear principal component analysis method is used for dimension reduction processing of the coal mine tunneling face gas data in Table 1, and the first three principal components of the data are calculated to achieve a cumulative contribution rate of more than 99%. For example Figure 3 .
[0065] Wherein F1-F6 are relative coal mine tunneling face gas emission, coal mine tunneling face gas pressure, desorption coefficient, coal mine tunneling face gas trend peak ratio, coal mine tunneling face gas trend peak average ratio, ground stress parameter and ground stress index, and the first three data are selected as the main data.
[0066] The test data after dimension reduction of the nuclear principal component analysis is imported into the POA-LSTM, PSO-LSTM and LSTM prediction model, and the prediction methods of LSTM, PSO-LSTM and POA-LSTM are compared;
[0067] From Figure 4It can be seen that the prediction error of the traditional long short-term memory network is large. Since the LSTM algorithm stores some information in the cell state and calculates new information through the Sigmoid function, abnormal data is generated in the cross calculation process due to the reliance on the bias vector of each structure in the calculation process. In 30 sample data, the maximum relative error of the LSTM algorithm is 16.05%, the average relative error is 9.6%, and the root mean square error value is 36.64; the maximum relative error rate of the coal mine tunneling working face gas prediction data obtained by the PSO-LSTM prediction model is 21.85%, the average relative error rate is 8.22%; the maximum relative error rate of the coal mine tunneling working face gas prediction data obtained by the POA-LSTM prediction model is 5.8%, the average relative error rate is 1.93%, and the root mean square error value is 5.324, as shown in Table 2 in detail.
[0068] Table 2 Comparison of prediction errors
[0069]
[0070]
[0071] Therefore, compared with the existing particle swarm long short-term memory network algorithm, the long short-term memory network. The accuracy of the algorithm coupled with the peacock optimization algorithm is greatly improved, and the adaptive calculation of the peacock optimization algorithm improves the convergence speed and avoids falling into local optimum.
[0072] The POA-LSTM algorithm effectively reduces the values of RMSE and R-square.
[0073] Therefore, it can be proved that the long short-term memory network optimized by the peacock optimization algorithm and kernel principal component analysis has greatly improved accuracy. According to the MATLAB 2016A software, the POA-LSTM, PSO-LSTM and LSTM prediction error result graph is shown in Figure 5.
[0074] According to the particle swarm optimization long short-term memory network model, the long short-term memory network model and the peacock optimization algorithm optimized long short-term memory network error result graph, Figure 5 As can be seen from the figure, the POA-LSTM prediction model is most coupled with the prediction result of the gas emission of the coal mine tunneling working face and the actual gas emission of the coal mine tunneling working face. The PSO-LSTM model is second, the LSTM model is farthest, and there are abnormal data values, indicating that the POA-LSTM prediction model is closest to the actual value.
[0075] In the three gas prediction models of the coal mine tunneling working face described above, the change range of the absolute error curve is: the change range of the LSTM coal mine tunneling working face gas prediction model is the largest, the change range of the PSO-LSTM coal mine tunneling working face gas prediction model is less than that of the POA-LSTM, and the reason is that the defect of falling into local optimization in the above two LSTM algorithms has not been solved. Because the local optimal value exists, it leads to a large error in the training process of the neural network; the PSO-LSTM model is unstable, although the overall error is smaller than that of the LSTM, which is because the ordinary LSTM cannot guarantee that the gradient disappearance problem in the training process can be effectively solved.
[0076] The specific embodiments of the application are described in detail above in combination with the drawings, but the application is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the purpose of the application.
Claims
1. A coal mine tunneling face gas emission amount prediction method based on a KPCA-POA-LSTM model, characterized by: The specific steps of the method are as follows: Step 1, using kernel principal component analysis KPCA to initialize the initial data, setting the radial basis kernel function parameter, and calculating the characteristic vector and characteristic value, and calculating the cumulative contribution rate; The initial data includes gas pressure, desorption coefficient, relative gas emission amount, gas trend peak ratio, gas trend peak average ratio, ground stress parameter or ground stress index of the coal mining working face; Step 2, constructing a POA-LSTM model, using peacock optimization algorithm POA to optimize the forget gate, input gate, input node, output gate, intermediate output and state unit of long short-term memory network, and reducing the gradient vanishing problem; Step 3, constructing a multi-dimensional state matrix and selecting an attractive strategy, using the optimized long short-term memory network to train the multi-dimensional state matrix, using Adam to solve according to the original data of the coal mining working face, and predicting the result of the coal mining working face; The specific steps of step 2 are as follows: Step 2.1, using the adaptive search and approach mechanism of female peacocks for optimization; the following mathematical model is used to describe the approach mechanism of female peacocks: （1） where is a random number uniformly distributed in ; is the position vector of the female peacock, is the current iteration number, is the maximum iteration number; and and are set to 0.1 and 1, respectively; , , , , are the five best individuals in the peacock population, according to formula (4), when the female peacock moves close to the selected peacock, which represents the local exploitation in the search process; when the female peacock tends to move to the symmetric position of the selected peacock, indicating global exploration in the search process; Step 2.2, using the adaptive search behavior of peacock chicks for optimization; the position update of the finder is described as follows: as follows: （2） （3） （4） where and represent two different one-dimensional random vectors uniformly distributed over and respectively; denotes a constant set to 1.5,! denotes the factorial; the specific behavior of each peacock offspring is modeled as: （5） （6） where represents a random number uniformly distributed on ; and denote the position vectors of the selected peacock and peacock cub, respectively; and denote two coefficient factors that dynamically change with the iteration number and are defined by the following equations: （7） （8） The above equations (7), (8) show that at the beginning of iteration greater than , the peacock offspring performs random search; when the iteration ends greater than , the peacock offspring converges to the optimal five solutions, that is, the optimal five peacock positions; , , and and equal to 0.1 and 1 respectively, the output parameters are optimized by the peacock search behavior.
2. The KPCA-POA-LSTM model-based coal mine tunneling face gas emission amount prediction method according to claim 1, characterized in that: The specific steps of step 1 are as follows: Step 1.
1. Construct the kernel principal component analysis, set the matrix X, sample points , where each column of X represents a sample Each sample point is a K-dimensional column vector, there are N samples in X, introduce a symmetric matrix K, and make its eigenvectors constitute the entire input space , where is the mapping, respectively represent the matrix space row and column; Step 1.2, Introduce kernel function, where is a low-dimensional vector, is a constant, map it to a high-dimensional vector using a radial basis function kernel Get the data norm and compute the eigenvectors of the matrix.