Ultrathin coal seam hydraulic support supporting quality evaluation method based on multi-index analysis
Through multi-index analysis and machine learning algorithms, the single index problem of the support quality evaluation of extremely thin coal seam hydraulic support is solved, and the comprehensive evaluation of hydraulic support is achieved, which improves the accuracy and reliability of the evaluation and avoids the cutting-top pressure frame. The accident occurred.
Patent Information
- Application Number
- CN202410911567.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-09
- Publication Date
- 2024-12-03
AI Technical Summary
In the prior art, the evaluation method of hydraulic support quality of extremely thin coal seams lacks a comprehensive evaluation of multiple indicators, which leads to a poor quality of support for a single bracket and can easily lead to a top-cutting rack accident. Moreover, the evaluation of thick coal seams is mostly a single indicator, and lacks a unified Comprehensive evaluation method.
The support quality evaluation method of extremely thin coal seam hydraulic support based on multi-index analysis is adopted. Through the steps of data acquisition, processing and identification, building matrix and chart, building model and input data, and adjusting real-time weights, combining support vector regression algorithm and XGBoost algorithm , construct a scoring group evaluation score and weight adjustment model to achieve comprehensive evaluation.
It can more accurately analyze the support quality of the hydraulic support, adjust the weight ratio in real time, and obtain more accurate evaluation results. It comprehensively evaluates the roof pressure of the bracket, the expansion and contraction of the left and right columns and other indicators, which improves the support quality evaluation of the hydraulic support under the coal seam. .
Smart Images

Figure CN119066529A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a coal seam hydraulic support support quality evaluation method, in particular to an ultra-thin coal seam hydraulic support support quality evaluation method based on multi-index analysis, and belongs to the technical field of coal mining. Background Art
[0002] With the decrease of thick and medium-thick coal seam reserves and the increasing scarcity of resources, the mining of extremely thin coal seams has attracted great attention from the country. At present, there are studies on the monitoring and evaluation methods of support quality in thick coal seams, but since extremely thin coal seams have only received attention in recent years, there is a lack of research on monitoring and evaluation methods of support quality.
[0003] Since the rock burst in very thin coal seams and thick coal seams may manifest differently, there will be some differences in data processing and analysis. Very thin coal seams are more likely to show local high stress concentration, while thick coal seams may show more widespread stress distribution. Among the geological characteristics of coal seams, very thin coal seams usually have higher ground stress and are prone to rock burst. The geological conditions are complex and the ground pressure changes more dramatically.
[0004] The monitoring of coal seam support mainly includes:
[0005] Monitoring indicators: The quality monitoring of hydraulic support for extremely thin coal seams needs to pay more attention to the support of the support for thin coal seams, including indicators such as support pressure and support deformation; while the quality monitoring of hydraulic support for thick coal seams may need to pay more attention to the support of the support for rock formations, including indicators such as rock formation deformation and rock pressure changes.
[0006] Monitoring technology: Due to the particularity of thin coal seams, the quality monitoring of hydraulic support in extremely thin coal seams may require the use of more sophisticated monitoring technologies, such as strain gauges, inclinometers, etc.; while the quality monitoring of hydraulic support in thick coal seams may use more traditional monitoring technologies, such as pressure gauges, displacement sensors, etc.
[0007] Monitoring frequency: Due to the variability of thin coal seams, the quality monitoring of hydraulic support in extremely thin coal seams may need to be carried out more frequently to promptly detect and deal with support problems; while the quality monitoring of hydraulic support in thick coal seams may adopt a more sparse monitoring frequency.
[0008] In the prior art, 1) a method for evaluating the quality of hydraulic support support based on the entropy weight TOPSIS method disclosed in the publication number CN115795739A includes the identification of the opening process of the safety valve, the identification of the leakage process of the hydraulic support, and the identification of the statistical period of the initial support force of the hydraulic support; extraction of characteristic process parameters of the hydraulic support support: including extraction of characteristic parameters of the opening of the safety valve, extraction of characteristic parameters of the leakage process of the hydraulic support, and extraction of characteristic parameters of the initial support process of the hydraulic support; forward processing of characteristic process parameters of the hydraulic support support; normalization of the forward matrix; calculation of the weights of each parameter of the hydraulic support by the entropy weight method; calculation of the evaluation score of each hydraulic support and comprehensive ranking; 2 ...2528229A The disclosed method for evaluating the quality of hydraulic support support of a working face based on control chart analysis collects working face mine pressure data of multiple hydraulic support cycle processes according to the hydraulic support cycle process, identifies the initial support force data in the working face mine pressure data collected each time, uses the initial support force data obtained each time as an initial support force subgroup sample, calculates the subgroup mean and subgroup standard deviation of each initial support force subgroup sample; monitors the subgroup mean and subgroup standard deviation of one or multiple consecutive initial support force subgroup samples through a pre-made reference Shewhart control chart; wherein the reference Shewhart control chart includes a control chart and an S control chart; and determines whether the quality of hydraulic support support in the hydraulic support cycle process meets the standard based on the judgment result of step S2. However, the current existing technology still has gaps in the evaluation method of the support quality of extremely thin coal seams: on the one hand, due to factors such as insufficient initial support force of the hydraulic support, aging of the sealing ring, poor cleanliness of the emulsion, frequent opening of the safety valve, unreasonable working posture of the support, and high intensity of periodic pressure, the support quality of the hydraulic support of the comprehensive mining working face will be attenuated to a certain extent. The deterioration of the support quality of a single support can easily cause an increase in the working resistance of the adjacent supports, and even cause severe accidents such as cutting the top and crushing the support. On the other hand, the evaluation methods for the support quality of thick coal seams are mostly single-indicator evaluations, and there are fewer multi-indicator evaluation methods. In addition, there is no definite evaluation relationship between the indicators, and there is no unified comprehensive evaluation of the evaluation results. Summary of the invention
[0009] The purpose of the present invention is to provide a method for evaluating the quality of hydraulic support for ultra-thin coal seams based on multi-indicator analysis in order to solve the problem that most of the evaluation methods for the quality of support are single-indicator evaluation, there are few multi-indicator rating methods, there is no definite evaluation relationship between the indicators, and there is no unified comprehensive evaluation of the evaluation results.
[0010] The present invention achieves the above-mentioned purpose through the following technical scheme: A method for evaluating the quality of ultra-thin coal seam hydraulic support support based on multi-index analysis, comprising the following steps:
[0011] Step 1: Data Collection
[0012] The hydraulic support electro-hydraulic control system is used to collect the real-time posture data of the hydraulic support. The collected data is the extension and contraction amount of the left and right columns of the hydraulic support.
[0013] Use pressure sensors placed on hydraulic support columns to collect real-time pressure data of each column, including initial support force data, spatiotemporal working resistance data, and end-of-work resistance data;
[0014] The real-time pressure data of each position of the top plate is obtained by using multiple sets of sensors arranged on the top plate of the hydraulic support;
[0015] Step 2: Data processing and identification
[0016] The obtained data of telescopic amount of left and right columns of hydraulic support, real-time pressure data of columns and real-time pressure data of each position of top plate are processed and converted into digital signals. After being transmitted to the processing center, the data are cleaned, digitally filtered and calibrated;
[0017] The processed data is identified, including identifying the real-time pressure data of the column according to the time and pressure value, obtaining the real-time pressure data of the hydraulic support column in each working cycle, and then identifying the pressure data of each working cycle to obtain the initial support force data of the column support group and the resistance data at the end of the working cycle;
[0018] The expansion and contraction amounts of the left and right columns of the hydraulic support are classified to obtain the spatiotemporal data information of the left column and the spatiotemporal data information of the right column;
[0019] Identify the pressure data of each position of the top plate of the hydraulic support to obtain the pressure data of each position of the top plate of the hydraulic support in each working cycle, and group the pressure data of each position of each hydraulic support into a group to obtain a top plate pressure data group of the hydraulic support group;
[0020] Step 3: Build a matrix and a chart
[0021] A real-time pressure matrix is constructed based on the column pressure data of the hydraulic support group on the working surface, a pressure vector is constructed based on the initial support force data within a single working cycle of the hydraulic support group, and a pressure vector is constructed based on the end-of-working-cycle resistance data within a single working cycle of the hydraulic support group;
[0022] Construct a real-time pressure chart based on the column pressure data of the hydraulic support group on the working surface;
[0023] According to the telescopic data of the columns of the hydraulic support group on the working surface, a real-time telescopic matrix of the left and right columns is constructed;
[0024] A real-time expansion difference matrix is constructed according to the expansion data difference of the left and right columns;
[0025] A real-time pressure data group is constructed according to the pressure data group of each position of the top plate of a single hydraulic support, and the pressure variance value of each hydraulic support top plate is calculated according to the constructed real-time pressure data group, and then the variance value is synthesized into a real-time pressure variance matrix about each top plate;
[0026] Step 4: Build model and input data
[0027] Construct a regression algorithm scoring training model for each indicator selected for the ultra-thin coal seam hydraulic support, input each matrix obtained in step 3 into the pre-trained initial data training model corresponding to each indicator, obtain the evaluation score of each indicator of the support group, the evaluation score and the indicator data corresponding to the optimal score of the current time period, and average the values corresponding to the optimal score except for the leakage indicator and the hydraulic valve working status indicator, and use the average as the scoring criterion for the data training model in the next time period;
[0028] Step 5: Adjust real-time weights
[0029] Construct an improved XGBoost algorithm weight adjustment model, set the initial weight ratio of each indicator to be evenly divided, input the data and matrix obtained in step 3 into the weight adjustment model, and obtain the optimal weight ratio of each indicator at present;
[0030] Step 6: Calculate scores and provide feedback on work performance
[0031] The real-time evaluation scores of each index of the support group obtained in step 4 and step 5 are multiplied by the real-time weight ratio to obtain the total score of the hydraulic support support quality, and the cause of the fault is obtained based on the specific low score.
[0032] As a further solution of the present invention: in step 3, the real-time pressure matrix constructed according to the column pressure data of the hydraulic support group on the working surface is:
[0033]
[0034] Where: n is the number of hydraulic supports on the working face, m is the number of time points taken, and P is obtained. ij It represents the column pressure value of the i-th hydraulic support at the j-th time point;
[0035] Construct the initial support force vector, [P1 P2 P3 … P n ], where P i It represents the initial support force of the i-th hydraulic support in this working cycle;
[0036] Construct the end-of-cycle resistance vector, [P1′ P2′ P3′ … P n ′], where P1′ represents the initial support force of the i-th hydraulic support in this working cycle;
[0037] A real-time pressure chart is constructed based on the column pressure data of the hydraulic support group on the working surface.
[0038] As a further solution of the present invention: in the step 3, the real-time telescopic amount matrix of the left and right columns constructed according to the telescopic amount data of the columns of the hydraulic support group on the working surface is:
[0039]
[0040] Where: n is the number of hydraulic supports on the working face, m is the number of time points taken, L ij represents the telescopic value of the left column of the i-th hydraulic support at the j-th time point, L′ ij It is the extension value of the right column.
[0041] As a further solution of the present invention: in step 3, the real-time expansion difference matrix constructed according to the expansion data difference of the left and right columns is:
[0042]
[0043] Where: n is the number of hydraulic supports on the working face, m is the number of time points taken, ΔL ij It represents the extension and contraction value of the column of the i-th hydraulic support at the j-th time point.
[0044] As a further solution of the present invention: in step 3, the pressure values of the top plate of the i-th hydraulic support are set to P1, P2, P3, ..., P m ;
[0045] Take the average,
[0046] According to the variance formula, The real-time pressure variance value matrix of each top plate is constructed as follows:
[0047]
[0048] in: It represents the variance value of the top plate pressure of the i-th hydraulic support at the j-th time point.
[0049] As a further solution of the present invention: the step 4 specifically includes:
[0050] S41. After determining the support vector regression algorithm, using the established relational model feature variables, SVR support vector regression is a special case of the SVM support vector machine algorithm. The hyperplane decision boundary in the SVM is the SVR regression model, and the specific expression of its basic form is as follows:
[0051] f(x)=w T x+b
[0052] Where: w is the normal vector, x i Belongs to n-dimensional vector, which is a sample data set, x = [x1, x2, x3…x n ]; b represents a constant term, which is represented by w T x+b=0 to obtain;
[0053] For the support vector machine regression method, if the sample point is close to the model, that is, the sample point is within the interval boundary of the model, then the sample does not consider the loss, and the corresponding loss function is called ε-insensitive loss:
[0054] L(z)=max(0,|z|-ε)
[0055] Where: y p This is the actual value of the sample at this moment, y r is the sample initial value, Z=y p -y r ; ε is a hyperparameter related to the width of the interval boundary, which is equivalent to f(x) = w T x+b is the center, and the width of the interval is 2ε. If the training sample is in the interval, then the prediction is considered accurate without any loss.
[0056] S42. Determine the stress capacity scoring formula. The explained variance is used to measure the model's ability to explain the fluctuations in the data set. The closer the score is to 1, the better the effect. The expression is as follows:
[0057]
[0058] Where: y t is the next moment value predicted by the SVR model, y p is the actual value of the sample, y rrs is the sample initial value, Var(y t -y p ) represents the residual, Var represents the variance, and then multiply EV by 100 to get the score of a single hydraulic support at a certain moment;
[0059] S43, score the column pressure, define the initial score as 100, and the initial values of P0 and λ as the sample actual value that explains the variance closest to the initial value;
[0060] in:
[0061]
[0062] Where: EV(P0) is the score corresponding to the initial value;
[0063] Input pressure data, process the data, obtain pressure matrix, initial support force vector, end-of-work resistance vector, and then divide the training set to enter the next step;
[0064] Transfer to the training model, train the parameters, complete the parameter training to obtain the column pressure score matrix, calculate the explained variance to obtain the total score of the column pressure of the bracket group, output the score matrix, the total score, and the number, time, and reason of the low-scoring bracket;
[0065] S44, scoring the column expansion, the process and model are consistent with the column pressure scoring;
[0066] S45, scoring the top plate pressure of the hydraulic support, the process and model are consistent with the column pressure scoring;
[0067] S46. Score the oil leakage of the bracket and the working status of the hydraulic valve, and construct a real-time pressure chart Pt according to the column pressure to observe whether there is any up and down fluctuation:
[0068] If there are multiple large value fluctuations in a working cycle, then the stent is leaking;
[0069] If the working resistance increases to a certain value in a working cycle and then fluctuates within a certain range, it means that the safety valve is working;
[0070] According to the Pt diagram, the difference between the maximum and minimum values of each fluctuation is calculated, and the oil leakage is evaluated according to the fluctuation size within a working cycle. The score is 100 when there is no waveform. Then, according to the fluctuation range and the fluctuation value, the score segment is set to obtain the leakage score.
[0071] Then, according to the Pt diagram, the working period of the safety valve is determined, and the working state of the safety valve is determined according to the fluctuation value ΔP and the safety valve's own parameters.
[0072] As a further solution of the present invention: in S43, the training of parameters specifically includes:
[0073] Substitute the processed data into the SVR regression model and loss function, and make a conditional judgment based on the initial value P0 obtained by the threshold predicted by the SVR model. If |P-P0|<λ, substitute the L(z) value obtained by the loss function into the explained variance, and get a score of 85 or more based on its value. Otherwise, make the next judgment, |P-P0|<(n+1)λ, and make one more judgment for every increase of n, and the result score will be reduced by 15 points.
[0074] Then, input it into the relative deviation to get the score. When it exceeds the initial value range, P0+(n+1)λ is used as the initial value of the deviation formula, and then the score is calculated. The specific formula of relative deviation is as follows:
[0075]
[0076] Where: RE represents relative deviation;
[0077] The bracket group scoring matrix and the reasons for low scores are obtained, and the pressures corresponding to brackets with scores above 85 are averaged as the initial values for the next moment.
[0078] As a further solution of the present invention: in the step 5, an improved XGBoost algorithm weight adjustment model is constructed, and the loss function measures the difference between a single prediction result and the actual result, and the loss function is as follows:
[0079]
[0080] Where: Loss is the sum of the loss function, which represents the sum of the model prediction error and the model complexity; α is the regularization coefficient of the tree complexity; T is the number of model leaf nodes, that is, the number of evaluation indicators; ω is the weight of each leaf node;
[0081] After training a single result, we get the value of the loss function, measure the accuracy of the model, and then draw the entire objective function;
[0082] The objective function is a comprehensive evaluation of the performance of the entire model, usually including the sum of the loss function and possible regularization terms. The objective function is as follows:
[0083]
[0084] Where: l is the loss function; For prediction The difference between y and y. In this model, y is the value of the i-th sample at the previous moment. is the current value; Ω is a regular term, which is used to reduce the complexity of the model and prevent overfitting;
[0085] T is the number of leaf nodes in the model, that is, the number of evaluation indicators, which represents the complexity of the model. The more indicators there are, the more complex the model is; γ is the regularization coefficient of the model complexity, which controls the weight of the tree complexity in the objective function; ω is the weight of the leaf node, that is, the weight of each indicator, which limits the overfitting phenomenon caused by excessive weight; λ is the L2 regularization coefficient of the weight.
[0086] As a further solution of the present invention: in the step 5, adjusting the weight ratio specifically includes:
[0087] Set the weight single step length to 0.01 and the minimum value of the objective function to L min, pre-process the data, input the support column pressure data, column expansion data and roof pressure data, and then divide the training set. Substitute the corresponding loss function according to the determined initial weight ratio, and make conditional judgment according to the obtained value. If L <L min , enter the next judgment, otherwise, increase the weight by one step and enter the loop; enter the next judgment, when all the obtained L are less than L min , the optimal weight ratio is obtained and the index weight ratio adjustment starts at the next moment.
[0088] The beneficial effects of the present invention are:
[0089] 1) The present invention adds two indicators, namely, the pressure at each position of the roof and the expansion and contraction of the left and right columns, to the existing indicators, which can more comprehensively evaluate the support quality of the hydraulic support under the coal seam;
[0090] 2) The present invention establishes a regression scoring model for each indicator, which can achieve a more accurate analysis of the data collected by the stent, and establishes a weight adjustment model based on the influence of each indicator on the comprehensive evaluation result, which can adjust the weight ratio in real time and obtain a more accurate evaluation result. BRIEF DESCRIPTION OF THE DRAWINGS
[0091] Figure 1 It is a schematic diagram of the steps of the present invention;
[0092] Figure 2 This is a schematic diagram of a single working cycle Pt of the present invention;
[0093] Figure 3 A schematic diagram of the modeling process of the support group column pressure regression feedback model of the present invention;
[0094] Figure 4 It is a schematic diagram of the modeling process of the bracket group index XBG weight adjustment model of the present invention. DETAILED DESCRIPTION
[0095] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0096] Embodiment 1, as Figures 1 to 4 As shown, a method for evaluating the quality of hydraulic support support in an ultra-thin coal seam based on multi-index analysis is provided. The method for evaluating the quality of hydraulic support support in an ultra-thin coal seam comprises the following steps:
[0097] Step 1: Data Collection
[0098] The hydraulic support electro-hydraulic control system of the intelligent mining working face is used to collect the real-time posture data of the hydraulic support. The collected data mainly includes the extension and retraction of the left and right columns of the hydraulic support.
[0099] Use the pressure sensors arranged on the hydraulic support columns to collect the real-time pressure data of each column, including initial support force data, spatiotemporal working resistance data and end-of-work resistance data;
[0100] The real-time pressure data of each position of the top plate is obtained by using multiple sets of sensors arranged on the top plate of the hydraulic support;
[0101] Step 2: Data processing and identification
[0102] The obtained data of telescopic amount of left and right columns of hydraulic support, real-time pressure data of columns, and pressure data of each position of top plate are processed and converted into digital signals. After being transmitted to the processing center, the data are cleaned, digitally filtered, and calibrated.
[0103] Then the processed data is identified, including identifying the real-time pressure data of the column according to the time and pressure values, obtaining the real-time pressure data of the hydraulic support column in each working cycle, and then identifying the pressure data of each working cycle to obtain the initial support force data of the column support group and the resistance data at the end of the working cycle;
[0104] Classify the expansion and contraction amounts of the left and right columns to obtain the spatiotemporal data information of the left column and the spatiotemporal data information of the right column;
[0105] Identify the pressure data of each position of the top plate of the hydraulic support to obtain the pressure data of each position of the top plate of the hydraulic support in each working cycle, and group the pressure data of each position of each hydraulic support into a group to obtain a top plate pressure data group of the hydraulic support group;
[0106] Step 3: Build a matrix and a chart
[0107] A real-time pressure matrix is constructed based on the column pressure data of the hydraulic support group on the working surface, a pressure vector is constructed based on the initial support force data within a single working cycle of the hydraulic support group, and a pressure vector is constructed based on the end-of-working-cycle resistance data within a single working cycle of the hydraulic support group;
[0108] Construct a real-time pressure chart based on the column pressure data of the hydraulic support group on the working surface;
[0109] According to the telescopic data of the columns of the hydraulic support group on the working surface, a real-time telescopic matrix of the left and right columns is constructed;
[0110] A real-time expansion difference matrix is constructed according to the expansion data difference of the left and right columns;
[0111] A real-time pressure data group is constructed according to the pressure data group of each position of the top plate of a single hydraulic support, and the pressure variance value of each hydraulic support top plate is calculated according to the constructed real-time pressure data group, and then the variance value is synthesized into a real-time pressure variance matrix about each top plate;
[0112] Step 4: Build model and input data
[0113] Construct a regression algorithm scoring training model for each indicator selected for the ultra-thin coal seam hydraulic support, input each matrix obtained in step 3 into the pre-trained initial data training model corresponding to each indicator, obtain the evaluation score of each indicator of the support group, the evaluation score and the indicator data corresponding to the optimal score of the current time period, and average the values corresponding to the optimal score except for the leakage indicator and the hydraulic valve working status indicator, and use the average as the scoring criterion for the data training model in the next time period;
[0114] Step 5: Adjust real-time weights
[0115] Construct an improved XGBoost algorithm weight adjustment model, set the initial weight ratio of each indicator to be evenly divided, that is, the initial weights of [column pressure score, top plate pressure score, column expansion score, column leakage score, hydraulic valve working status score] are [0.25, 0.25, 0.25, 0.25], input the data and matrix obtained in step 3 into the weight adjustment model, and obtain the optimal weight ratio of each indicator at present;
[0116] Step 6: Calculate scores and provide feedback on work performance
[0117] The real-time evaluation scores of each index of the support group obtained in step 4 and step 5 are multiplied by the real-time weight ratio to obtain the total score of the hydraulic support support quality, and the cause of the fault is obtained based on the specific low score.
[0118] Embodiment 2, in addition to all the technical features of Embodiment 1, this embodiment also includes: In step 3, the real-time pressure matrix constructed according to the column pressure data of the hydraulic support group on the working surface is:
[0119]
[0120] Among them, n is the number of hydraulic supports on the working face, m is the number of time points taken, and P is obtained. ij It represents the column pressure value of the i-th hydraulic support at the j-th time point;
[0121] Construct the initial support force vector, [P1 P2 P3 … P n ], where P i It represents the initial support force of the i-th hydraulic support in this working cycle;
[0122] Construct the end-of-cycle resistance vector, [P1′ P2′ P3′ … P n ′], where P1′ represents the initial support force of the i-th hydraulic support in this working cycle;
[0123] A real-time pressure chart is constructed based on the column pressure data of the hydraulic support group on the working surface.
[0124] Furthermore, the real-time telescopic amount matrix of the left and right columns constructed according to the telescopic amount data of the columns of the hydraulic support group on the working surface is:
[0125]
[0126] Where: n is the number of hydraulic supports on the working face, m is the number of time points taken, L ij represents the telescopic value of the left column of the i-th hydraulic support at the j-th time point, L′ ij It is the extension value of the right column.
[0127] Furthermore, the real-time expansion difference matrix constructed according to the expansion data difference of the left and right columns is:
[0128]
[0129] Where n is the number of hydraulic supports on the working face, m is the number of time points taken, ΔL ij It represents the extension and contraction value of the column of the i-th hydraulic support at the j-th time point.
[0130] Furthermore, suppose the pressure values of the top plate of the i-th hydraulic support are P1, P2, P3, ..., P m , take the average,
[0131] According to the variance formula, The real-time pressure variance value matrix of each top plate is constructed as follows:
[0132]
[0133] in: represents the variance value of the top plate pressure of the i-th hydraulic support at the j-th time point.
[0134] Embodiment 3, in addition to all the technical features of Embodiment 1, this embodiment also includes: Step 4, specifically including:
[0135] S41. After determining the support vector regression algorithm, using the established relational model feature variables, SVR support vector regression is a special case of the SVM support vector machine algorithm. The hyperplane decision boundary in the SVM is the SVR regression model, and the specific expression of its basic form is as follows:
[0136] f(x)=w T x+b
[0137] Where: w is the normal vector, x i Belongs to n-dimensional vector, which is a sample data set, x = [x1, x2, x3…x n ]; b represents a constant term, which is represented by w T x+b=0 to obtain;
[0138] For the support vector machine regression method, if the sample point is close to the model, that is, the sample point is within the interval boundary of the model, then the sample does not consider the loss, and the corresponding loss function is called ε-insensitive loss:
[0139] L(z)=max(0,|z|-ε)
[0140] Where: y p This is the actual value of the sample at this moment, y r is the sample initial value, Z=y p -y r ; ε is a hyperparameter related to the width of the interval boundary, which is equivalent to f(x) = w T x+b is the center, and the width of the interval is 2ε. If the training sample is in the interval, then the prediction is considered accurate without any loss.
[0141] S42. Determine the stress capacity scoring formula. The explained variance is used to measure the model's ability to explain the fluctuations in the data set. The closer the score is to 1, the better the effect. The expression is as follows:
[0142]
[0143] Where: y t is the next moment value predicted by the SVR model, y p is the actual value of the sample, y rrs is the sample initial value, Va r (y t -y p ) represents the residual, Var represents the variance, and then multiply EV by 100 to get the score of a single hydraulic support at a certain moment;
[0144] S43, score the column pressure, define the initial score as 100, and the initial values of P0 and λ as the sample actual value that is closest to the explained variance at the initial time;
[0145] in:
[0146]
[0147] Where: EV(P0) is the score corresponding to the initial value;
[0148] Input pressure data, process the data, obtain pressure matrix, initial support force vector, end-of-work resistance vector, and then divide the training set to enter the next step;
[0149] Transfer to the training model, train the parameters, complete the parameter training to obtain the column pressure score matrix, calculate the explained variance to obtain the total score of the column pressure of the bracket group, output the score matrix, the total score, and the number, time, and reason of the low-scoring bracket;
[0150] S44, scoring the column expansion, the process and model are consistent with the column pressure scoring;
[0151] S45, scoring the top plate pressure of the hydraulic support, the process and model are consistent with the column pressure scoring;
[0152] S46. Score the oil leakage of the bracket and the working status of the hydraulic valve. According to the real-time pressure chart (Pt chart) constructed by the column pressure, observe whether there are up and down fluctuations. If there are multiple large value fluctuations in a working cycle, then the bracket has leakage; if in a working cycle, when the working resistance increases to a certain value, the working resistance fluctuates within a certain range, it means that the safety valve is working;
[0153] According to the Pt diagram, the difference between the maximum and minimum values of each fluctuation is calculated, and the oil leakage is evaluated according to the fluctuation size within a working cycle; the score is 100 when there is no waveform, and then according to the fluctuation range and the fluctuation value, the score segment is set to obtain the leakage score;
[0154] Then, according to the Pt diagram, the working period of the safety valve is determined, and the working state of the safety valve is determined according to the fluctuation value ΔP and the safety valve's own parameters.
[0155] Furthermore, the training of parameters specifically includes:
[0156] Substitute the processed data into the SVR regression model and loss function, and make a conditional judgment based on the initial value P0 obtained by the threshold predicted by the SVR model. If |P-P0|<λ, substitute the L(z) value obtained by the loss function into the explained variance, and get a score of 85 or more based on its value. Otherwise, make the next judgment, |P-P0|<(n+1)λ, and make one more judgment for every increase of n, and the result score will be reduced by 15 points.
[0157] Then, input it into the relative deviation to get the score. When it exceeds the initial value range, P0+(n+1)λ is used as the initial value of the deviation formula, and then the score is calculated. The specific formula of relative deviation is as follows:
[0158]
[0159] Where: RE represents relative deviation;
[0160] The bracket group scoring matrix and the reasons for low scores are obtained, and the pressures corresponding to brackets with scores above 85 are averaged as the initial values for the next moment.
[0161] Embodiment 4, in addition to all the technical features in Embodiment 1, this embodiment also includes: in step 5, an improved XGBoost algorithm weight adjustment model is constructed, and the loss function measures the difference between a single prediction result and the actual result, and its loss function is as follows:
[0162]
[0163] Where: Loss is the sum of the loss function, which represents the sum of the model prediction error and the model complexity; α is the regularization coefficient of the tree complexity; T is the number of model leaf nodes, that is, the number of evaluation indicators; ω is the weight of each leaf node;
[0164] After training a single result, we get the value of the loss function, measure the accuracy of the model, and then draw the entire objective function;
[0165] The objective function is a comprehensive evaluation of the performance of the entire model, usually including the sum of the loss function and possible regularization terms. The objective function is as follows:
[0166]
[0167] Where: l is the loss function; For prediction The difference between y and y. In this model, y is the value of the i-th sample at the previous moment. is the current value; Ω is a regular term, which is used to reduce the complexity of the model and prevent overfitting;
[0168] T is the number of leaf nodes in the model, that is, the number of evaluation indicators, which represents the complexity of the model. The more indicators there are, the more complex the model is; γ is the regularization coefficient of the model complexity, which controls the weight of the tree complexity in the objective function; ω is the weight of the leaf node, that is, the weight of each indicator, which limits the overfitting phenomenon caused by excessive weight; λ is the L2 regularization coefficient of the weight.
[0169] Furthermore, in step 5, the weight ratio is adjusted, the weight single step length is set to 0.01, and the minimum value L of the objective function is set to min , pre-process the data, input the support column pressure data, column expansion data and roof pressure data, and then divide the training set. Substitute the corresponding loss function according to the determined initial weight ratio, and make conditional judgment according to the obtained value. If L <L min, enter the next judgment, otherwise, increase the weight by one step and enter the loop; enter the next judgment, when all the obtained L are less than L min , the optimal weight ratio is obtained and the index weight ratio adjustment starts at the next moment.
[0170] On the basis of existing indicators, the pressure at each position of the roof and the expansion and contraction of the left and right columns were added to more comprehensively evaluate the support quality of the hydraulic support under the coal seam. A regression scoring model was established to analyze the data collected by the support more accurately, and a weight adjustment model was established based on the influence of each indicator on the comprehensive evaluation result. The weight ratio can be adjusted in real time to obtain more accurate evaluation results.
[0171] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.
[0172] In addition, it should be understood that although the present specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.
Claims
1. A method for evaluating the quality of hydraulic support support in ultra-thin coal seams based on multi-index analysis, characterized in that: The method for evaluating the quality of ultra-thin coal seam hydraulic support support comprises the following steps: Step 1: Data Collection The hydraulic support electro-hydraulic control system is used to collect the real-time posture data of the hydraulic support. The collected data is the extension and contraction amount of the left and right columns of the hydraulic support. Use pressure sensors placed on hydraulic support columns to collect real-time pressure data of each column, including initial support force data, spatiotemporal working resistance data, and end-of-work resistance data; The real-time pressure data of each position of the top plate is obtained by using multiple sets of sensors arranged on the top plate of the hydraulic support; Step 2: Data processing and identification The obtained data of telescopic amount of left and right columns of hydraulic support, real-time pressure data of columns and real-time pressure data of each position of top plate are processed and converted into digital signals. After being transmitted to the processing center, the data are cleaned, digitally filtered and calibrated; The processed data is identified, including identifying the real-time pressure data of the column according to the time and pressure value, obtaining the real-time pressure data of the hydraulic support column in each working cycle, and then identifying the pressure data of each working cycle to obtain the initial support force data of the column support group and the resistance data at the end of the working cycle; The expansion and contraction amounts of the left and right columns of the hydraulic support are classified to obtain the spatiotemporal data information of the left column and the spatiotemporal data information of the right column; Identify the pressure data of each position of the top plate of the hydraulic support to obtain the pressure data of each position of the top plate of the hydraulic support in each working cycle, and group the pressure data of each position of each hydraulic support into a group to obtain a top plate pressure data group of the hydraulic support group; Step 3: Build a matrix and a chart A real-time pressure matrix is constructed based on the column pressure data of the hydraulic support group on the working surface, a pressure vector is constructed based on the initial support force data within a single working cycle of the hydraulic support group, and a pressure vector is constructed based on the end-of-working-cycle resistance data within a single working cycle of the hydraulic support group; Construct a real-time pressure chart based on the column pressure data of the hydraulic support group on the working surface; According to the telescopic data of the columns of the hydraulic support group on the working surface, a real-time telescopic matrix of the left and right columns is constructed; A real-time expansion difference matrix is constructed according to the expansion data difference of the left and right columns; A real-time pressure data group is constructed according to the pressure data group of each position of the top plate of a single hydraulic support, and the pressure variance value of each hydraulic support top plate is calculated according to the constructed real-time pressure data group, and then the variance value is synthesized into a real-time pressure variance matrix about each top plate; Step 4: Build model and input data Construct a regression algorithm scoring training model for each indicator selected for the ultra-thin coal seam hydraulic support, input each matrix obtained in step 3 into the pre-trained initial data training model corresponding to each indicator, obtain the evaluation score of each indicator of the support group, the evaluation score and the indicator data corresponding to the optimal score of the current time period, and average the values corresponding to the optimal score except for the leakage indicator and the hydraulic valve working status indicator, and use the average as the scoring criterion for the data training model in the next time period; Step 5: Adjust real-time weights Construct an improved XGBoost algorithm weight adjustment model, set the initial weight ratio of each indicator to be evenly divided, input the data and matrix obtained in step 3 into the weight adjustment model, and obtain the optimal weight ratio of each indicator at present; Step 6: Calculate scores and provide feedback on work performance The real-time evaluation scores of each index of the support group obtained in step 4 and step 5 are multiplied by the real-time weight ratio to obtain the total score of the hydraulic support support quality, and the cause of the fault is obtained based on the specific low score.
2. The method for evaluating the quality of ultra-thin coal seam hydraulic support support according to claim 1 is characterized in that: In step 3, the real-time pressure matrix constructed based on the column pressure data of the hydraulic support group on the working surface is: Where: n is the number of hydraulic supports on the working face, m is the number of time points taken, P ij It represents the column pressure value of the i-th hydraulic support at the j-th time point; Construct the initial support force vector, [P1 P2 P3…P n ], where P i It represents the initial support force of the i-th hydraulic support in this working cycle; Construct the end-of-cycle resistance vector, [P1′ P2′ P3′…P n ′], where P1′ represents the initial support force of the i-th hydraulic support in this working cycle; A real-time pressure chart is constructed based on the column pressure data of the hydraulic support group on the working surface.
3. The method for evaluating the quality of ultra-thin coal seam hydraulic support support according to claim 1 is characterized in that: In the step 3, the real-time telescopic amount matrix of the left and right columns constructed according to the telescopic amount data of the columns of the hydraulic support group on the working surface is: Where: n is the number of hydraulic supports on the working face, m is the number of time points taken, L ij represents the extension value of the left column of the i-th hydraulic support at the j-th time point, L ij It is the extension value of the right column.
4. The method for evaluating the quality of ultra-thin coal seam hydraulic support support according to claim 1 is characterized in that: In step 3, the real-time expansion difference matrix constructed according to the expansion data difference of the left and right columns is: Where: n is the number of hydraulic supports on the working face, m is the number of time points taken, ΔL ij It represents the extension and contraction value of the column of the i-th hydraulic support at the j-th time point.
5. The method for evaluating the quality of ultra-thin coal seam hydraulic support support according to claim 4 is characterized in that: In step 3, the pressure values of the top plate of the i-th hydraulic support are set to P1, P2, P3, ..., P m , take the average, According to the variance formula: The real-time pressure variance value matrix of each top plate is constructed as follows: in: It represents the variance value of the top plate pressure of the i-th hydraulic support at the j-th time point.
6. The method for evaluating the quality of ultra-thin coal seam hydraulic support support based on multi-index analysis according to claim 1 is characterized in that: The step 4 specifically includes: S41. After determining the support vector regression algorithm, using the established relational model feature variables, SVR support vector regression is a special case of the SVM support vector machine algorithm. The hyperplane decision boundary in the SVM is the SVR regression model, and the specific expression of its basic form is as follows: f(x)=w T x+b Where: w is the normal vector, x i Belongs to n-dimensional vector, which is a sample data set, x = [x1, x2, x3…x n ]; b represents the constant top, which is represented by w T x+b=0 to obtain; For the support vector machine regression method, if the sample point is close to the model, that is, the sample point is within the interval boundary of the model, then the sample does not consider the loss, and the corresponding loss function is called ε-insensitive loss: L(z)=max(0,|z|-ε) Where: y p This is the actual value of the sample at this moment, y r is the sample initial value, Z=y p -y r ; ε is a hyperparameter related to the width of the interval boundary, which is equivalent to f(x) = w T x+b is the center, and the width of the interval is 2ε. If the training sample is in the interval, then the prediction is considered accurate without any loss. S42. Determine the stress capacity scoring formula. The explained variance is used to measure the model's ability to explain the fluctuations in the data set. The closer the score is to 1, the better the effect. The expression is as follows: Where: y t is the next moment value predicted by the SVR model, y p is the actual value of the sample, y rrs is the sample initial value, Var(y t -y p ) represents the residual, Var represents the variance, and then multiply EV by 100 to get the score of a single hydraulic support at a certain moment; S43, score the column pressure, define the initial score as 100, and the initial values of P0 and λ as the sample actual value that explains the variance closest to the initial value; in: Where: EV(P0) is the score corresponding to the initial value; Input pressure data, process the data, obtain pressure matrix, initial support force vector, end-of-work resistance vector, and then divide the training set to enter the next step; Transfer to the training model, train the parameters, complete the parameter training to obtain the column pressure score matrix, calculate the explained variance to obtain the total score of the column pressure of the bracket group, output the score matrix, the total score, and the number, time, and reason of the low-scoring bracket; S44, scoring the column expansion, the process and model are consistent with the column pressure scoring; S45, scoring the top plate pressure of the hydraulic support, the process and model are consistent with the column pressure scoring; S46. Score the oil leakage of the bracket and the working state of the hydraulic valve. According to the real-time pressure chart Pt constructed by the column pressure, observe whether there is up and down fluctuation. If there are multiple large value fluctuations in a working cycle, then the bracket has leakage; if in a working cycle, when the working resistance increases to a certain value, the working resistance fluctuates within a certain range, it means that the safety valve is working; According to the Pt diagram, the difference between the maximum and minimum values of each fluctuation is calculated, and the oil leakage is evaluated according to the fluctuation size within a working cycle. The score is 100 when there is no waveform. Then, according to the fluctuation range and the fluctuation value, the score segment is set to obtain the leakage score. Then, according to the Pt diagram, the working period of the safety valve is determined, and the working state of the safety valve is determined according to the fluctuation value ΔP and the safety valve's own parameters.
7. The method for evaluating the quality of ultra-thin coal seam hydraulic support support according to claim 6 is characterized in that: In S43, the training of parameters specifically includes: Substitute the processed data into the SVR regression model and loss function, and make a conditional judgment based on the initial value P0 obtained by the threshold predicted by the SVR model. If |P-P0|<λ, substitute the L(z) value obtained by the loss function into the explained variance, and get a score of 85 or more based on its value. Otherwise, make the next judgment, |P-P0|<(n+1)λ, and make one more judgment for every increase of n, and the result score will be reduced by 15 points. Then, input it into the relative deviation to get the score. When it exceeds the initial value range, P0+(n+1)λ is used as the initial value of the deviation formula, and then the score is calculated. The specific formula of relative deviation is as follows: Where: RE represents relative deviation; The bracket group scoring matrix and the reasons for low scores are obtained, and the pressures corresponding to brackets with scores above 85 are averaged as the initial values for the next moment.
8. The method for evaluating the quality of ultra-thin coal seam hydraulic support support according to claim 1 is characterized by: In the step five, Construct an improved XGBoost algorithm weight adjustment model. The loss function measures the difference between a single predicted result and the actual result. The loss function is as follows: Where: Loss is the sum of the loss function, which represents the sum of the model prediction error and the model complexity; α is the regularization coefficient of the tree complexity; T is the number of model leaf nodes, that is, the number of evaluation indicators; ω is the weight of each leaf node; After training a single result, we get the value of the loss function, measure the accuracy of the model, and then draw the entire objective function; The objective function is a comprehensive evaluation of the performance of the entire model, usually including the sum of the loss function and possible regularization terms. The objective function is as follows: Where: l is the loss function; For prediction The degree of difference between y and y. In this model, y is the value of the i-th sample at the previous moment. is the current value; Ω is a regular term, which is used to reduce the complexity of the model and prevent overfitting; T is the number of leaf nodes in the model, that is, the number of evaluation indicators, which represents the complexity of the model. The more indicators there are, the more complex the model is; γ is the regularization coefficient of the model complexity, which controls the weight of the tree complexity in the objective function; ω is the weight of the leaf node, that is, the weight of each indicator, which limits the overfitting phenomenon caused by excessive weight; λ is the L2 regularization coefficient of the weight.
9. The method for evaluating the quality of ultra-thin coal seam hydraulic support support according to claim 1 is characterized in that: In the step 5, adjusting the weight ratio specifically includes: Set the weight single step length to 0.01 and the minimum value of the objective function to L min , pre-process the data, input the support column pressure data, column expansion data and roof pressure data, and then divide the training set. Substitute the corresponding loss function according to the determined initial weight ratio, and make conditional judgment according to the obtained value. If L<L min , enter the next judgment, otherwise, increase the weight by one step and enter the loop; enter the next judgment, when all the obtained L are less than L min , the optimal weight ratio is obtained and the index weight ratio adjustment starts at the next moment.
Citation Information
Patent Citations
Working face hydraulic support supporting quality evaluation method based on control chart analysis
CN112528229A
Hydraulic support supporting quality evaluation method based on entropy weight TOPSIS method
CN115795739A
