A method for evaluating the health status of coal mining machines

Abstract

The present invention provides a method for evaluating the health status of a coal mining machine, comprising the following steps: calculating the comprehensive correlation coefficient between each monitoring parameter in the collected coal mining machine status data, and taking the monitoring parameters below the preset threshold as the indicators participating in the evaluation; obtaining the static weight vectors of each indicator according to the hierarchical analysis method and the entropy weight method respectively, obtaining the dynamic weight vector by the variable weight formula, and obtaining the evaluation vector Q in combination with the gray clustering; inputting the monitoring parameter data to be evaluated into the trained XGBoost model to obtain the evaluation probability vector R of the coal mining machine status of the XGBoost model; weightedly fusing the evaluation vector Q and the evaluation probability vector R to obtain the comprehensive evaluation vector QR; and obtaining the probability size corresponding to each coal mining machine status according to the comprehensive evaluation vector QR. The method takes into account the influence of the coal mining machine working condition factors and the operation data in the evaluation process, combines the empirical model with the artificial intelligence model, mutually weakens the use restrictions of the two algorithms, and enhances the rationality of the evaluation results.

Description

Technical Field

[0001] The invention relates to the technical field of equipment maintenance, and in particular to a method for evaluating the health status of a coal mining machine. Background Art

[0002] Intelligent coal mining is a new stage in the development of comprehensive mechanization of coal, among which intelligent coal mining is a key step in the development of intelligent coal mining. As the core equipment of coal mining, the reliability and safety of coal mining machines are important prerequisites for the normal operation of comprehensive mining, ensuring the safety of coal production, and improving the economic benefits of coal enterprises. The working environment of coal mining machines is complex and changeable. Harsh factors such as heavy load operation, frequent startup, and narrow operating space make it difficult to monitor its status and frequent failures. Therefore, closely combining with coal mining conditions, carrying out active perception of coal mining machine status, studying coal mining machine data processing, indicator analysis, health status assessment and other methods are important means to achieve full life cycle tracking, health management and services for coal mining machines, and are the basic guarantee for achieving safe, efficient, green and intelligent coal mining.

[0003] The existing health status assessment of coal mining machines mainly has the following four problems: (1) It is difficult to detect abnormal data of coal mining machines, and the selection of evaluation indicators is not reasonable. (2) Empirical evaluation algorithms such as hierarchical analysis method and fuzzy comprehensive evaluation are affected by the structure and working conditions of the coal mining machine itself, and human intervention is inevitable in the evaluation process. At the same time, relevant research on evaluation indicator construction, indicator weight allocation, and health status classification is still insufficient. (3) Common artificial intelligence evaluation methods such as support vector machines have difficulties in solving multi-classification problems. Models such as deep learning and neural networks are easily limited by sample data. It is difficult to have a reasonable explanation for the evaluation model and results, and the model generalization ability is poor. (4) The evaluation model established based on the empirical model or artificial intelligence evaluation model alone is difficult to take into account the influence of coal mining machine working conditions and operating data, and the evaluation results are difficult to conform to the actual working conditions. The empirical model considers a large number of evaluation factors, but the information provided by the coal mining machine operating data is not studied in depth. The intelligent evaluation model analyzes the deep meaning expressed by the coal mining machine operating data, but it is difficult to consider the influence of actual experience and evaluation factors.

[0004] Therefore, the present invention proposes a new method for evaluating the health status of coal mining machines. Summary of the invention

[0005] To solve the above problems, the present invention proposes a coal mining machine health status assessment method that integrates variable weight fuzzy comprehensive evaluation and XGBoost to address the limitations of a single assessment algorithm. The results of the two methods, VW-FCE and XGBoost, are integrated to obtain the final result from the multi-model assessment results. This method takes into account the influence of coal mining machine operating factors and operating data during the assessment process, combines empirical models with artificial intelligence models, weakens the use limitations of the two algorithms, and enhances the rationality of the assessment results.

[0006] To achieve the above object, the present invention provides the following technical solutions.

[0007] A method for evaluating the health status of a coal mining machine comprises the following steps:

[0008] The comprehensive correlation coefficients between the various monitoring parameters in the collected coal mining machine status data are calculated, and the monitoring parameters below the preset threshold are used as indicators for evaluation;

[0009] The static weight vector of each indicator is obtained according to the hierarchical analysis method and the entropy weight method respectively, and the dynamic weight vector is obtained by the variable weight formula. The dynamic weight vector is combined with the grey clustering to obtain the evaluation vector Q;

[0010] Input the monitoring parameter data to be evaluated into the trained XGBoost model, calculate the leaf node score corresponding to each state of the coal mining machine, and obtain the evaluation probability vector R of the coal mining machine state of the XGBoost model;

[0011] The evaluation vector Q and the evaluation probability vector R are weightedly fused to obtain a comprehensive evaluation vector QR;

[0012] The probability size corresponding to each coal mining machine state is obtained according to the comprehensive evaluation vector QR, and the coal mining machine state result is obtained according to the maximum membership degree.

[0013] Preferably, the method further includes: performing abnormal data detection on the collected coal mining machine status data through an isolation forest algorithm, and removing the abnormal data.

[0014] Preferably, the calculation of the comprehensive correlation coefficient between the various monitoring parameters in the collected coal mining machine status data includes the following steps:

[0015] The Pearson correlation coefficient, Spearman correlation coefficient and cosine distance were used to analyze the correlation of various monitoring parameters of the coal mining machine:

[0016]

[0017]

[0018]

[0019] Where: p , ρ s , ρ c are the Pearson correlation coefficient, Spearman correlation coefficient, and cosine distance between parameters respectively; X, Y are sample vectors, Cov(X,Y) is the covariance of X and Y, Var(X) and Var(Y) are the variances of X and Y; a is the number of samples, N is the total number of samples, d a It represents the difference in the position of the two variables after they are sorted.

[0020] Comprehensive correlation coefficient ρ co for:

[0021] ρ co =α1×ρ s +α2×ρ p +α3×ρ c

[0022] In the formula, α1, α2, α3 are weights and α1+α2+α3=1; the three correlation coefficients have equal importance, that is, α1=α2=α3=1 / 3;

[0023] When the comprehensive correlation coefficient is greater than or equal to the threshold TH, the parameters are closely related, and when it is less than the threshold TH, the parameters are not closely related;

[0024] According to the correlation coefficient ρ co Based on the calculation results, the monitoring parameters with comprehensive correlation coefficients lower than TH were screened out to complete the construction of evaluation indicators.

[0025] Preferably, the step of obtaining a dynamic weight vector from a static weight vector by a variable weight formula comprises the following steps:

[0026] Determine the number of indicators involved in the evaluation m;

[0027] According to the hierarchical analysis method and entropy weight method, the j-th indicator weight vector u is obtained respectively j 、v j , combine the two weight vectors to obtain the j-th indicator combination weight:

[0028]

[0029] The combined weight vector is obtained as:

[0030] w=(w1,w2,...,w m )

[0031] The variable weight formula is determined as:

[0032]

[0033] Where: is the variable weight coefficient of the jth indicator, w j is the combined weight coefficient, x j Running parameters for the jth indicator;

[0034] Introduce the equilibrium function into the variable weight formula, and the equilibrium function is:

[0035]

[0036] Substitute the equilibrium function into the variable weight formula, then:

[0037]

[0038] Where: j is the variable weight coefficient of the jth indicator after the introduction of the equilibrium function; according to the variable weight formula of the introduction of the equilibrium function, the final variable weight, that is, the dynamic weight vector, is obtained as follows:

[0039] ω=[ω1,ω2,...,ω m ]

[0040] Among them, α is 0.5.

[0041] Preferably, the step of combining the dynamic weight vector with grey clustering to obtain the evaluation vector Q comprises the following steps:

[0042] The health status of the coal mining machine is evaluated through the central point type triangular whitening weight function, and m indicators involved in the evaluation are obtained, and their membership degree is:

[0043]

[0044] Where: s is the number of gray classes, λ t is the center of the tth gray class, x j is the monitoring value of a certain indicator;

[0045] Get x j The conformity vector for s gray classes is:

[0046]

[0047] Calculate the conformity vector of the monitoring values ​​of all indicators to s gray classes, and get the conformity matrix as

[0048] F=[F1,F2,...,F m ]

[0049] Construct the conformity matrix F for s gray classes. According to the dynamic weight vector, the evaluation vector of each state of the coal mining machine is obtained as follows:

[0050] Q T =Fω T .

[0051] Preferably, the acquisition of the evaluation probability vector R of the coal mining machine state of the XGBoost model includes the following steps:

[0052] Construct sample data of monitoring parameters of coal mining machines in various states, and use XGBoost model to train the sample data;

[0053] Input the monitoring parameter data to be evaluated into the trained XGBoost model and calculate the leaf node score corresponding to each state of the coal mining machine;

[0054] Add the scores P of a certain state obtained under each tree to get the total score S of each state of the sample to be evaluated O for:

[0055]

[0056] Where: k is the number of trees, O is the state level, P(O) k is the score corresponding to different states in each tree, then there is a scoring matrix S:

[0057] S=[S1,S2,...,S s ]

[0058] Find the sigmoid for each score, get the evaluation probability corresponding to each state, and finally get the XGBoost probability vector, that is, the evaluation probability vector R:

[0059] R=[R1,R2,...,R s ]

[0060] Where: s is the health status level of the coal mining machine, R1, R2, ..., R s is the evaluation result under this sample, that is, the probability of each state.

[0061] Preferably, the acquisition of the comprehensive evaluation vector QR comprises the following steps:

[0062] The evaluation vector Q and the evaluation probability vector R are weighted and fused to obtain the final comprehensive evaluation vector QR:

[0063] QR=[QR1,QR2,...,QR s ]

[0064] in

[0065] QR O =(β1Q O +β2R O )(O=1,2,...,s)

[0066] Where: QR Ois the comprehensive evaluation probability of each state, β1 and β2 represent the probability weights of the two algorithms respectively; among them, β1 and β2 are 0.5.

[0067] Beneficial effects of the present invention:

[0068] (1) This method uses the isolation forest algorithm to complete the detection of abnormal data of coal mining machines, which can effectively solve the problem of high difficulty in detecting abnormal data of coal mining machines.

[0069] (2) This method proposes a method for calculating the comprehensive correlation coefficient of monitoring parameters, completes the construction of coal mining machine evaluation indicators, and can solve the problem of unreasonable selection of coal mining machine evaluation indicators.

[0070] (3) The variable weight fuzzy comprehensive evaluation proposed in this method can provide a better explanation for the indicator weight allocation, the establishment of the algorithm model and the evaluation results. Under the combination of subjective and objective factors, the model is more applicable when training data is missing.

[0071] (4) Aiming at the influence of coal mining machine working conditions and operation data, the variable weight fuzzy comprehensive evaluation method and the advantages of XGBoost algorithm are combined to weightedly fuse the evaluation vectors of the two to obtain a comprehensive evaluation vector, and the final result is obtained from the evaluation results of the empirical model and the artificial intelligence model. This method weakens the use restrictions of the two algorithms and enhances the rationality of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of evaluation based on variable weight fuzzy comprehensive evaluation and XGBoost method according to an embodiment of the present invention;

[0073] Figure 2 This is a graph of the detection results of an isolated forest abnormal data according to an embodiment of the present invention;

[0074] Figure 3 is a correlation coefficient heat map of an embodiment of the present invention;

[0075] Figure 4 is a curve diagram of dynamic weight changes of indicators in an embodiment of the present invention;

[0076] Figure 5 is a relationship diagram of the evaluation accuracy of different parameter models in an embodiment of the present invention;

[0077] Figure 6 is a confusion matrix of evaluation results of an embodiment of the present invention;

[0078] Figure 7 It is a curve diagram of the health status change of the coal mining machine according to an embodiment of the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0080] Example 1

[0081] The present invention proposes a method for evaluating the health status of a coal mining machine. The method flow chart is as follows: Figure 1 As shown, the specific steps include:

[0082] S1: The collected coal mining machine status data is detected and removed through the isolation forest algorithm. The isolation forest algorithm is specifically suitable for high-dimensional data sets and has good detection accuracy for global outliers, data with strong mutation and large uncertainty.

[0083] S2: Calculate the comprehensive correlation coefficient between each monitoring parameter in the collected coal mining machine status data, and use the monitoring parameters below the preset threshold as indicators for participating in the evaluation.

[0084] Specific:

[0085] The Pearson correlation coefficient, Spearman correlation coefficient and cosine distance were used to analyze the correlation of various monitoring parameters of the coal mining machine:

[0086]

[0087]

[0088]

[0089] Where: p , ρ s , ρ c are the Pearson correlation coefficient, Spearman correlation coefficient, and cosine distance between parameters respectively; X, Y are sample vectors, Cov(X,Y) is the covariance of X and Y, Var(X) and Var(Y) are the variances of X and Y; a is the number of samples, N is the total number of samples, d a It represents the difference in the position of the two variables after they are sorted.

[0090] Comprehensive correlation coefficient ρ co for:

[0091] ρ co =α1×ρ s +α2×ρ p +α3×ρ c

[0092] In the formula, α1, α2, α3 are weights and α1+α2+α3=1; the three correlation coefficients have equal importance, that is, α1=α2=α3=1 / 3;

[0093] When the comprehensive correlation coefficient is greater than or equal to the threshold TH, the parameters are closely related, and when it is less than the threshold TH, the parameters are not closely related;

[0094] According to the correlation coefficient ρ co Based on the calculation results, the monitoring parameters with comprehensive correlation coefficients lower than TH were screened out to complete the construction of evaluation indicators.

[0095] S3: Obtain the static weight vector of each indicator according to the hierarchical analysis method and the entropy weight method respectively, obtain the dynamic weight vector by the variable weight formula, and combine the dynamic weight vector with grey clustering to obtain the evaluation vector Q.

[0096] S3.1: Acquisition of dynamic weight vector:

[0097] Determine the number of indicators involved in the evaluation m;

[0098] According to the hierarchical analysis method and entropy weight method, the j-th indicator weight vector u is obtained respectively j 、v j , combine the two weight vectors to obtain the j-th indicator combination weight:

[0099]

[0100] The combined weight vector is obtained as:

[0101] w=(w1,w2,...,w m )

[0102] The variable weight formula is determined as:

[0103]

[0104] Where: is the variable weight coefficient of the jth indicator, w j is the combined weight coefficient, x j Running parameters for the jth indicator;

[0105] Introduce the equilibrium function into the variable weight formula, and the equilibrium function is:

[0106]

[0107] Substitute the equilibrium function into the variable weight formula, then:

[0108]

[0109] Where: jis the variable weight coefficient of the jth indicator after the introduction of the equilibrium function; according to the variable weight formula of the introduction of the equilibrium function, the final variable weight, that is, the dynamic weight vector, is obtained as follows:

[0110] ω=[ω1,ω2,...,ω m ]

[0111] Among them, α is 0.5.

[0112] S3.2: Acquisition of evaluation vector Q:

[0113] The health status of the coal mining machine is evaluated through the central point type triangular whitening weight function, and m indicators involved in the evaluation are obtained, and their membership degree is:

[0114]

[0115] Where: s is the number of gray classes, λ t is the center of the tth gray class, x j is the monitoring value of a certain indicator;

[0116] Get x j The conformity vector for s gray classes is:

[0117]

[0118] Calculate the conformity vector of the monitoring values ​​of all indicators to s gray classes, and get the conformity matrix as

[0119] F=[F1,F2,...,F m ]

[0120] Construct the conformity matrix F for s gray classes. According to the dynamic weight vector, the evaluation vector of each state of the coal mining machine is obtained as follows:

[0121] Q T =Fω T .

[0122] S4: Input the monitoring parameter data to be evaluated into the trained XGBoost model, calculate the leaf node score corresponding to each state of the coal mining machine, and obtain the evaluation probability vector R of the coal mining machine state of the XGBoost model.

[0123] S4.1: Prediction model construction:

[0124] The prediction model of XGBoost is

[0125]

[0126] Where: x i is the feature of the i-th sample, f k (x i) is the predicted value of the kth tree, is the predicted value of the model. The XGBoost algorithm starts with the first tree and optimizes sequentially until the Kth tree is completed. For a given x i , then the model is initialized:

[0127]

[0128] Add the Kth tree to the model:

[0129]

[0130] The objective function of XGBoost consists of two parts: the loss function that measures the training error and the regularization term that controls the complexity. The formula is:

[0131]

[0132] Where: is the loss function, which is used to measure the error between the true value and the model prediction value. The commonly used loss function in the classification model is the mean square error MSE; and Ω is the regularization term, which is used to control the complexity of the model to prevent the model from overfitting. Ω(f k ) represents the complexity of the kth tree, and c is a constant term.

[0133] The loss function is expanded according to the Taylor series to obtain the approximate objective function, and the constant term c can be ignored.

[0134]

[0135] Where: g i is the loss function The first derivative of i express The second derivative of , where:

[0136]

[0137]

[0138] In the XGBoost classification model, the complexity is mainly determined by the number of leaf nodes and the smoothness of the corresponding node weights.

[0139]

[0140] Where: γ and All parameters are manually set, T is the number of leaf nodes, P is the weight of each leaf, that is, the leaf node score, is the regularization penalty term for the P weight parameter.

[0141] Use I rRepresents the sample set in the rth leaf, let:

[0142]

[0143] Find the optimal leaf node weight And the objective function obj * for:

[0144]

[0145]

[0146] S4.2: Obtaining the evaluation probability vector R:

[0147] Construct sample data of monitoring parameters of coal mining machines in various states, and use XGBoost model to train the sample data;

[0148] Input the monitoring parameter data to be evaluated into the trained XGBoost model and calculate the leaf node score corresponding to each state of the coal mining machine;

[0149] Add the scores P of a certain state obtained under each tree to get the total score S of each state of the sample to be evaluated O for:

[0150]

[0151] Where: k is the number of trees, O is the state level, P(O) k is the score corresponding to different states in each tree, then there is a scoring matrix S:

[0152] S=[S1,S2,...,S s ]

[0153] Find the sigmoid for each score, get the evaluation probability corresponding to each state, and finally get the XGBoost probability vector, that is, the evaluation probability vector R:

[0154] R=[R1,R2,...,R s ]

[0155] Where: s is the health status level of the coal mining machine, R1, R2, ..., R s is the evaluation result under this sample, that is, the probability of each state.

[0156] S5: Perform weighted fusion on the evaluation vector Q and the evaluation probability vector R to obtain a comprehensive evaluation vector QR.

[0157] Specific:

[0158] The evaluation vector Q and the evaluation probability vector R are weighted and fused to obtain the final comprehensive evaluation vector QR:

[0159] QR=[QR1,QR2,...,QR s ]

[0160] in

[0161] QR O =(β1Q O +β2R O )(O=1,2,...,s)

[0162] Where: QR O is the comprehensive evaluation probability of each state, β1 and β2 represent the probability weights of the two algorithms respectively; among them, β1 and β2 are 0.5.

[0163] S6: Obtain the probability size corresponding to each coal mining machine state according to the comprehensive evaluation vector QR, and obtain the coal mining machine state result according to the maximum membership degree.

[0164] Embodiment 1:

[0165] 1.1 Outlier Detection

[0166] Taking the two main operating parameters of the coal mining machine measured under working conditions, the main inverter current and the main inverter temperature as examples, the isolation forest algorithm is used to detect the abnormal values ​​of the coal mining machine parameters. There are 2000 groups of measurement data of the two parameters, of which there are 50 groups of abnormal data, mainly composed of abnormal signals such as coal mining machine start and stop, mutation, etc.

[0167] According to the requirements of the Isolation Forest Algorithm, the algorithm parameters are determined, and the number of samples n_samples is set to 2000, the number of trees n_estimators is set to 100, the outlier fraction of the data set is set to 0.025, and the outlier ratio of the data set is set to 0.025. Figure 2 shown.

[0168] In the figure, the isolation forest divides normal points into high-density groups, and the abnormal points are scattered in the space outside the high-density area. By isolating the abnormal points, a higher detection accuracy is obtained. Compared with the commonly used DBScan clustering and local outlier factor algorithm (LOF) two density-based outlier removal algorithms, the results are shown in Table 1.

[0169] Table 1 Comparison of abnormal data detection algorithms for coal mining machines

[0170]

[0171]

[0172] It can be seen from Table 1 that the isolation forest algorithm has higher recognition accuracy and shorter recognition time for abnormal data of coal mining machines.

[0173] 1.2 Evaluation Index Construction

[0174] Taking the monitoring data of coal mining machines under complex geological conditions as an example, the correlation of coal mining machine monitoring parameters is analyzed. The main monitoring parameters of coal mining machines are determined in combination with actual working conditions as shown in Table 2.

[0175] Table 2 Main monitoring parameters of coal mining machine

[0176]

[0177] The three-month monitoring data of the coal mining machine was selected, and the factors such as coal mining machine shutdown and emergency were considered. The data of 2000 time points that can characterize the operation status of the coal mining machine were extracted to form an 18×2000-dimensional data set. The comprehensive correlation coefficient was calculated and the correlation coefficient heat map was obtained as shown in the figure. Figure 3 shown.

[0178] Depend on Figure 3 It is known that the four parameters of the main inverter of the coal mining machine are closely related to the four parameters of the slave inverter, the temperature of the traction motor is closely related to the temperature of the right cutting motor, and the temperature of the pump motor is closely related to the temperature of the left and right cutting motors. Set the threshold TH to 0.9, and select the monitoring indicators with a correlation coefficient lower than 0.9 according to the comprehensive correlation coefficient to obtain the final evaluation indicators of the coal mining machine. The results are shown in Table 3.

[0179] Table 3 Selection of evaluation indicators for coal mining machines

[0180]

[0181] 2.1 Evaluation Method Training and Implementation

[0182] Sample data set creation

[0183] The correlation analysis of the monitoring parameters of the coal mining machine under working conditions was carried out, and finally 12-dimensional indicators were obtained to participate in the health assessment of the coal mining machine. The results are shown in Table 3. According to the determined evaluation indicators and some actual monitoring data, based on the Weibull distribution, a total of 2000 simulation data of the coal mining machine operation status were obtained, including 1000 training data, 300, 380, 250, 50, and 20 healthy, good, general, degraded, and fault data respectively; 1000 test data, 300, 380, 250, 50, and 20 healthy, good, general, degraded, and fault data respectively.

[0184] Model training and validation

[0185] (1) Variable weight fuzzy comprehensive evaluation

[0186] According to the analytic hierarchy process, the 12-dimensional evaluation indicators of coal mining machines (Table 3) are manually compared and scored, the discriminant matrix is ​​determined, the subjective weight is obtained based on the judgment matrix, and the matrix consistency test is completed. According to the entropy weight method, based on the training data, the objective weight is obtained through the information entropy of each indicator. The subjective weight is combined with the objective weight to obtain the combined weight, that is, the static weight vector as shown in Table 4.

[0187] Table 4 Coal mining machine index weight vector

[0188]

[0189] Then we can get the variation law of coal mining machine index weight coefficient with monitoring time. The results are as follows: Figure 4 shown.

[0190] Depend on Figure 7 It is known that when the coal mining machine works normally, the weight of the index is relatively stable. As the health of the coal mining machine deteriorates, the weight of each index shows different changes. When the coal mining machine has a preliminary failure, some original secondary indicators (such as pump motor temperature) gradually increase in weight and become more important indicators. Therefore, the dynamic weight can more accurately reflect the real operating status of each index of the coal mining machine than the static weight coefficient.

[0191] According to previous research and expert experience, the health status of coal mining machines is divided into O = {1, 2, 3, 4, 5} = {healthy, good, general, deteriorated, faulty}, and the corresponding whitening weight function range is determined. The health status level and the corresponding whitening weight function are shown in Table 5.

[0192] Table 5 Health status level and corresponding whitening weight function

[0193]

[0194] The health status of the coal mining machine was evaluated based on the variable weight fuzzy comprehensive evaluation method, and the results are shown in Table 6.

[0195] Table 6 Evaluation results of variable weight fuzzy comprehensive evaluation

[0196]

[0197]

[0198] According to the statistical results, the overall assessment accuracy of the health status of coal mining machines using this method is 92.4%.

[0199] (2) XGBoost Implementation

[0200] Based on the XGBoost algorithm, the processed coal mining machine data is used as the feature input, and the five health states of the coal mining machine are used as the classification output results. Set the XGBoost shrinkage step size eta to 0.1, the random sampling ratio subsample to 0.2, the random column ratio colsample_bytree to 1, the minimum loss function drop value gamma to 0, and the number of classification categories num_class to 5. Set the initial minimum leaf node weight and min_child_weight to 5. Select the maximum tree depth max_depth as 1, 2, 4, 6, set 1, 3, 5, 7, 9 as the min_child_weight value for model training, and get the relationship between each parameter and evaluation accuracy as shown below: Figure 5 a, b, c, d.

[0201] Depend on Figure 5 It can be seen that when the model max_depth is 2, it is more stable and has a higher evaluation accuracy than when max_depth is 1, 4, and 6. Figure 5 (b) It can be seen that as num_boost_round increases, the model min_child_weight is different, and the evaluation accuracy changes differently. When min_child_weight increases, the model accuracy shows a trend of first increasing and then decreasing or flattening. When min_child_weight = 7 and num_boost_round = 80, the model average accuracy is the highest.

[0202] The optimal parameter model is used to evaluate the health status of the coal mining machine. The evaluation results are shown in Table 7. The evaluation results of the five states of health, good, general, deterioration, and failure are statistically analyzed. Figure 6 shown.

[0203] Table 7 XGBoost evaluation results

[0204]

[0205] Depend on Figure 6 It can be seen that the coal mining machine condition assessment model based on XGBoost still has a high recognition accuracy for data with uneven sample distribution, and still has a certain assessment accuracy when there are insufficient degradation and fault samples. The overall assessment accuracy is 97.8%.

[0206] (3) VW-FCE-XGBoost evaluation model verification

[0207] The three methods of VW-FCE, XGBoost and VW-FCE-XGBoost are used to calculate the health status evaluation vector of the coal mining machine at each moment, and the health status change curve of the coal mining machine during operation is obtained as follows: Figure 7 shown.

[0208] Depend on Figure 7 It can be seen that all three methods can evaluate the health status of the equipment more accurately. At the same time, it can be seen that the health status is difficult to evaluate when it is in the alternation of two states (circled in the figure). The VW-FCE method shows good fuzziness when the probability values ​​of the two states are the same (at the intersection), but it is difficult to make a clear judgment on the state at that moment, and the reliability of the results obtained is not high. XGBoost made a timely judgment during the alternation between the two states, but ignored the changes in the health status during the transition period, and the judgment was too absolute. From VW-FCE-XGBoost, it can be seen that the combined results of the two methods can accurately judge the state at that moment, and can also express the transition process of the health state to a certain extent, and the results obtained are more interpretable.

[0209] VW-FCE-XGBoost was used to evaluate the health status of the coal mining machine. The accuracy of the method for evaluating the five health status of the coal mining machine, namely healthy, good, general, deteriorated, and faulty, was 100%, 98.7%, 99.2%, 100%, and 80%, respectively, and the overall evaluation accuracy of the model was 98.9%. Compared with the variable weight fuzzy comprehensive evaluation and XGBoost methods, the evaluation accuracy of the VW-FCE-XGBoost model increased by 6.5% and 1.1%.

[0210] Embodiment 2:

[0211] Taking the MG400 / 930-WD model coal mining machine in operation in a certain mine as an example, the coal mining machine is powered by a single cable with a power supply voltage of 3300 volts. The overall structure is a horizontal arrangement of multiple motors, and the traction method is airborne AC variable frequency speed regulation-pin track chainless traction.

[0212] The three-month operation monitoring data of the coal mining machine was selected, and the experimental data was subjected to outlier removal and index correlation analysis based on the isolated forest and comprehensive correlation coefficient, and the evaluation indicators were determined as shown in Table 3. The degradation trend of the coal mining machine is complex, and most of the faults are sudden faults and accidental faults. The unclear definition of health and fault data makes it difficult for the evaluation results to meet the actual working conditions. At the same time, different standards for determining faults and degradation will also lead to differences in the evaluation results. The coal mining machine data was processed and analyzed, and combined with the coal mining machine inspection records, the coal mining machine health status was finally divided into three states: O = {1, 2, 3} = {healthy, other, fault}, and 1000 training and test data were obtained, and the ratio of health, other, and fault data was 16:3:1. The evaluation results of the coal mining machine health status based on VW-FCE-XGBoost are shown in Table 8.

[0213] Table 8 Results of health status assessment of coal mining machines

[0214]

[0215] As shown in Table 8, the VW-FCE-XGBoost-based coal mining machine health status assessment method has an accuracy of 92.9%, 83.3%, and 90.0% for the three states of coal mining machine health, other, and fault, respectively. In the experiment, the characteristics of each health data are roughly the same; the characteristics of the fault data are diverse, but due to limited conditions, this paper selects a relatively single fault data; other data are highly uncertain, including single indicator degradation, multiple indicator degradation, start-stop signals, standby signals, and other data features. Therefore, the evaluation accuracy of this type of data is the worst. In summary, the experimental results meet the actual situation and have a certain degree of accuracy.

[0216] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for evaluating the health status of a coal mining machine, characterized in that: The following steps are involved: The comprehensive correlation coefficients between the various monitoring parameters in the collected coal mining machine status data are calculated, and the monitoring parameters below the preset threshold are used as indicators for evaluation; The static weight vector of each indicator is obtained according to the hierarchical analysis method and the entropy weight method respectively, and the dynamic weight vector is obtained by the variable weight formula. The dynamic weight vector is combined with the grey clustering to obtain the evaluation vector Q; Input the monitoring parameter data to be evaluated into the trained XGBoost model, calculate the leaf node score corresponding to each state of the coal mining machine, and obtain the evaluation probability vector R of the coal mining machine state of the XGBoost model; The evaluation vector Q and the evaluation probability vector R are weightedly fused to obtain a comprehensive evaluation vector QR; The probability size corresponding to each coal mining machine state is obtained according to the comprehensive evaluation vector QR, and the coal mining machine state result is obtained according to the maximum membership degree; The acquisition of the evaluation probability vector R of the coal mining machine state of the XGBoost model includes the following steps: Construct sample data of monitoring parameters of coal mining machines in various states, and use XGBoost model to train the sample data; Input the monitoring parameter data to be evaluated into the trained XGBoost model and calculate the leaf node score corresponding to each state of the coal mining machine; Add the scores P of a certain state obtained under each tree to get the total score S of each state of the sample to be evaluated O for: Where: k is the number of trees, O is the state level, P(O) k is the score corresponding to different states in each tree, then there is a scoring matrix S: S=[S1,S2,...,S s ] Find the sigmoid for each score, get the evaluation probability corresponding to each state, and finally get the XGBoost probability vector, that is, the evaluation probability vector R: R=[R1,R2,...,R s ] Where: s is the health status level of the coal mining machine, R1, R2, ..., R s is the evaluation result under this sample, that is, the probability of each state.

2. The method for evaluating the health status of a coal mining machine according to claim 1, characterized in that: Also includes: The isolated forest algorithm is used to detect abnormal data in the collected coal mining machine status data and eliminate abnormal data.

3. The method for evaluating the health status of a coal mining machine according to claim 1, characterized in that: The method of calculating the comprehensive correlation coefficient between the various monitoring parameters in the collected coal mining machine status data includes the following steps: The Pearson correlation coefficient, Spearman correlation coefficient and cosine distance were used to analyze the correlation of various monitoring parameters of the coal mining machine: Where: p , ρ s , ρ c are the Pearson correlation coefficient, Spearman correlation coefficient, and cosine distance between parameters respectively; X, Y are sample vectors, Cov(X,Y) is the covariance of X and Y, Var(X) and Var(Y) are the variances of X and Y; a is the number of samples, N is the total number of samples, d a It represents the difference in the position of the two variables after they are sorted. Comprehensive correlation coefficient ρ co for: r co =α1×ρ s +a2×p p +a3×r c In the formula, α1, α2, α3 are weights and α1+α2+α3=1; the three correlation coefficients have equal importance, that is, α1=α2=α3=1 / 3; When the comprehensive correlation coefficient is greater than or equal to the threshold TH, the parameters are closely related, and when it is less than the threshold TH, the parameters are not closely related; According to the correlation coefficient ρ co Based on the calculation results, the monitoring parameters with comprehensive correlation coefficients lower than TH were screened out to complete the construction of evaluation indicators.

4. The method for evaluating the health status of a coal mining machine according to claim 1, characterized in that: The step of obtaining a dynamic weight vector from a static weight vector by a variable weight formula comprises the following steps: Determine the number of indicators involved in the evaluation m; According to the hierarchical analysis method and entropy weight method, the j-th indicator weight vector u is obtained respectively j 、v j , combine the two weight vectors to obtain the j-th indicator combination weight: The combined weight vector is obtained as: in=(in1,in2,...,in m ) The variable weight formula is determined as: Where: is the variable weight coefficient of the jth indicator, w j is the combined weight coefficient, x j Running parameters for the jth indicator; Introduce the equilibrium function into the variable weight formula, and the equilibrium function is: Substitute the equilibrium function into the variable weight formula, then: Where: j is the variable weight coefficient of the jth indicator after the introduction of the equilibrium function; according to the variable weight formula of the introduction of the equilibrium function, the final variable weight, that is, the dynamic weight vector, is obtained as follows: ω=[ω1,ω2,...,ω m ] Among them, α is 0.

5.

5. The method for evaluating the health status of a coal mining machine according to claim 4, characterized in that: The step of combining the dynamic weight vector with the grey clustering to obtain the evaluation vector Q comprises the following steps: The health status of the coal mining machine is evaluated through the central point type triangular whitening weight function, and m indicators involved in the evaluation are obtained, and their membership degree is: Where: s is the number of gray classes, λ t is the center of the tth gray class, x j is the monitoring value of a certain indicator; Get x j The conformity vector for s gray classes is: Calculate the conformity vector of the monitoring values ​​of all indicators to s gray classes, and get the conformity matrix as F=[F1,F2,...,F m ] Construct the conformity matrix F for s gray classes. According to the dynamic weight vector, the evaluation vector of each state of the coal mining machine is obtained as follows: Q T =Fω T 。 6. The method for evaluating the health status of a coal mining machine according to claim 1, characterized in that: The acquisition of the comprehensive evaluation vector QR comprises the following steps: The evaluation vector Q and the evaluation probability vector R are weighted and fused to obtain the final comprehensive evaluation vector QR: QR=[QR1,QR2,...,QR s ] in QR O =(β1Q O +β2R O )(O=1,2,...,s) Where: QR O is the comprehensive evaluation probability of each state, β1 and β2 represent the probability weights of the two algorithms respectively; among them, β1 and β2 are 0.5.

Images