Support motion feature classification method based on pressure data of hydraulic support of fully mechanized coal mining face

Abstract

The invention relates to a fully mechanized coal face hydraulic support pressure data-based support motion feature classification method, and belongs to the field of fully mechanized coal face coal cutting feature classification. According to the method, original pressure data are smoothed through wavelet denoising, and the data are divided into three stages of initial supporting, working and unloading by using a change point detection technology. Then, extracting features such as an average value, a standard deviation and entropy of each stage, performing unsupervised learning by using a Gaussian mixture model, and classifying each stage into a primary supporting stage, a working stage or an unloading stage; according to the method, a large amount of manual marking training data is not needed, high-precision support motion state information can be provided under complex and changeable underground conditions, abnormal conditions can be found in time, safety accidents are prevented, the number of coal cutting circulation times is automatically counted, the working mode of the hydraulic support is optimized, and the production efficiency is improved.

Description

Technical Field

[0001] The invention belongs to the field of coal cutting feature classification in a fully mechanized mining face, and relates to a support motion feature classification method based on hydraulic support pressure data in a fully mechanized mining working face. Background Art

[0002] In the modern coal mining process, hydraulic support is one of the indispensable key equipment for fully mechanized mining face. It is not only used to support the roof and control the mine pressure, but also to ensure the safety and stability of the working space. With the continuous improvement of coal mine safety production standards, regulatory authorities have paid more and more attention to the real-time monitoring and intelligent analysis of the operating status of hydraulic supports. Traditional monitoring methods often rely on manual observation and recording, which is not only inefficient, but also easily affected by human factors, resulting in inaccurate or delayed data.

[0003] By accurately identifying the different movement characteristics of hydraulic supports, not only can abnormal situations be discovered in a timely manner and potential safety accidents be effectively prevented, but the number of coal cutting cycles can also be automatically counted, thereby avoiding common errors, omissions or data falsification problems in manual records. In addition, based on a deep understanding of these movement characteristics, the working mode of the hydraulic support can be further optimized, its response speed and adaptability can be improved, and the production efficiency of the entire fully mechanized mining face can be improved.

[0004] In order to solve the above problems, the present invention proposes a support motion feature classification model based on a machine learning algorithm. The model can automatically identify the motion characteristics of the hydraulic support by performing unsupervised analysis on the pressure data of the hydraulic support of the fully mechanized working face. There are three main types: initial support (i.e., the pressure value rises rapidly), working (the pressure value is in a flat state, accompanied by some fluctuations), and unloading (the pressure value suddenly rises and then drops rapidly). This automated method does not require a large amount of manual annotation of training data, and can provide high-precision support motion state information under complex and changeable underground conditions. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a method for classifying the motion characteristics of hydraulic supports based on the pressure data of hydraulic supports in fully mechanized mining working faces. It can solve the problem of automatic identification of the movement stages of hydraulic supports during coal mining, including the initial support stage (rapid pressure rise), the working stage (stable pressure with small fluctuations) and the unloading stage (sudden pressure rises and then drops rapidly). Taking into account the volatility of the data, especially in the working stage, powerful mathematical and computational techniques are used to smooth the signal, segment the time series, extract meaningful features and classify the stages. This automatic identification aims to improve the safety and efficiency of fully mechanized mining working faces by monitoring the pressure data of hydraulic supports in real time. It avoids the need for manual data annotation, helps to discover potential safety hazards, and improves operating efficiency by optimizing the coal cutting cycle and identifying anomalies in equipment operation.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A method for classifying support motion characteristics based on hydraulic support pressure data of a fully mechanized mining working face, the method comprising the following steps:

[0008] Perform wavelet denoising on the pressure data of hydraulic supports to smooth the original signal;

[0009] Using change point detection technology to segment the smoothed pressure data into different stages, the stages include an initial support stage, a working stage, and an unloading stage;

[0010] Extracting features for each stage, the features including mean, standard deviation and entropy;

[0011] Using unsupervised learning methods, each phase is classified into the initial support phase, the working phase, or the unloading phase according to the characteristics.

[0012] Furthermore, the unsupervised learning method is a Gaussian mixture model.

[0013] Furthermore, the change point detection technology is a pruning exact linear time (PELT) algorithm.

[0014] Further, the wavelet denoising comprises the following steps:

[0015] Perform continuous wavelet transform on time series data, convert it into wavelet domain, and obtain wavelet coefficients;

[0016] In the wavelet domain, threshold processing is applied to each coefficient, and the wavelet coefficients with absolute values ​​less than the threshold are set to zero;

[0017] The signal is reconstructed by inverse wavelet transform using the wavelet coefficients after threshold processing to obtain the denoised signal.

[0018] Further, the feature extraction for each stage comprises the following steps:

[0019] Calculate the average value for each stage;

[0020] Calculate the standard deviation for each stage;

[0021] Calculate the entropy at each stage.

[0022] Further, the classification of each stage into an initial support stage, a working stage or an unloading stage according to the characteristics specifically comprises the following steps:

[0023] Perform probability calculation on the feature vector of each stage to obtain the probability of each stage belonging to each Gaussian component;

[0024] Based on the probability calculation results, each stage is classified as the initial support stage, working stage or unloading stage.

[0025] Furthermore, in the Gaussian mixture model, each Gaussian component corresponds to a stage of the support movement.

[0026] The beneficial effects of the present invention are:

[0027] (1) By real-time monitoring of the pressure data of the hydraulic support and automatically identifying the movement characteristics of the support, abnormal situations can be discovered in time, potential safety accidents can be effectively prevented, and the safety of underground workers can be guaranteed.

[0028] (2) Automatically count the number of coal cutting cycles to avoid common errors, omissions or data falsification problems in manual records, thereby improving data accuracy and statistical efficiency.

[0029] (3) Based on a deep understanding of the support movement characteristics, the working mode of the hydraulic support can be optimized, its response speed and adaptability can be improved, thereby improving the production efficiency of the entire comprehensive mining face.

[0030] (4) There is no need for a large amount of manual annotation of training data, which reduces labor costs and training costs.

[0031] (5) The method has a simple process and is easy to operate. It does not require complicated equipment and technology and is suitable for on-site application in coal mines.

[0032] (6) This method is applicable to various types of hydraulic supports and can be used in different coal mine environments.

[0033] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0035] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0036] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0037] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0038] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0039] Figure 1 The flowchart of the present invention is shown below. The method combines signal processing, unsupervised learning and time series analysis to create an automatic intelligent system for identifying the operating phase of hydraulic supports in coal mine environments. Each step of the process is explained below:

[0040] 1. Wavelet denoising for data smoothing preprocessing

[0041] Wavelet denoising is used to smooth the original signal to reduce fluctuations that may mislead subsequent steps, especially when calculating derivatives or detecting phase changes. This preprocessing helps reduce noise while preserving the overall trend in the data, which can effectively handle frequent fluctuations that appear in the flat phase. By performing a continuous wavelet transform on the time series data x(t), it is converted to the wavelet domain and the wavelet coefficients W are obtained. x (a,b), the formula is as follows:

[0042]

[0043] Where a represents the scale parameter, which is related to frequency, and b represents the translation parameter, which is related to time.

[0044] Then in the wavelet domain, for each coefficient W x (a,b) Apply threshold processing. A threshold N is selected and the wavelet coefficients whose absolute values ​​are less than the threshold are set to zero to remove the parts that are considered to be noise. The wavelet coefficients after threshold processing are used to reconstruct the signal through inverse wavelet transform, thus obtaining the denoised signal x denoise (t).

[0045] 2. Stage segmentation through change point detection

[0046] To simplify subsequent writing, let the denoised data x denoise (t) = x′(t).

[0047] Change point detection (CPD) is used on x′(t) to segment the time series into different phases. This phase identifies points in the signal where statistical properties (such as mean or variance) change significantly, corresponding to transitions between phases.

[0048] The segmentation problem is defined as minimizing the following cost function:

[0049]

[0050] Among them, τ i is the detected change point (the change point that divides the time series into segments), C(·) is the cost function (used to measure the variability within the segment), m is the total number of change points or the total number of segments minus 1, β is a trade-off parameter used to prevent the penalty term of over-segmentation (larger β values ​​will lead to fewer change points, and the β value usually needs to be determined by continuous trial and error combined with the change points of the numerical curve graph). The output of this step is a set of change points {τ 1 ,τ 2 ,…,τ n}, which is used to divide the time series into multiple segments.

[0051] The stent pressure usually fluctuates greatly at different stages, so the present invention uses a variance-based cost function to detect the segmentation, and its calculation formula is as follows:

[0052]

[0053] in refers to τ i ~τ i+1 The variance in this segment.

[0054] In order to minimize the overall cost, we need to find a set of optimal change points τ 1 ,τ 2 ,…,τ m , to minimize the segmentation cost plus a penalty term βm.

[0055] The solution to this problem is solved using dynamic programming. The most effective algorithm to achieve this goal is the Pruning Exact Linear Time (PELT) algorithm, which checks potential change points in sequence and prunes suboptimal choices at the same time, and can find the optimal set of change points with a time complexity of O(T).

[0056] 3. Feature extraction

[0057] Through step 2, multiple value intervals are identified, each of which corresponds to a stage in the system (initial support, working, unloading). For each segment, the characteristics that characterize the dynamics of the stage are extracted. i ,τ i+1 ] as an example, the following features are extracted:

[0058] Average value μ i :

[0059]

[0060] Standard deviation i (Used to measure the degree of pressure fluctuation within a segment):

[0061]

[0062] Entropy H i (Measuring the signal complexity at this stage helps distinguish between stable and chaotic stages):

[0063]

[0064] The above extracted features are organized into feature vectors as follows:

[0065]

[0066] 4. Classification

[0067] Complete feature extraction and form the feature vector f for each segment i , the similarities or differences between the segments can be measured in order to group them into clusters. The present invention uses unsupervised learning techniques to classify each part into one of three stages: initial support stage, working or unloading.

[0068] Initial support stage: The initial stage when the hydraulic support supports the top plate, the pressure value will rise rapidly at this time.

[0069] Working stage: The stage when the hydraulic support supports the top plate and maintains stability. At this time, the pressure value will be in a flat state with some fluctuations.

[0070] Unloading stage: The stage when the hydraulic support unloads the top plate, the pressure value will suddenly increase and then decrease rapidly.

[0071] Since the time of each cycle of the stent motion is non-uniform and the motion time of each stage is also non-uniform, a Gaussian mixture model is used to classify the segments, which assumes that the data is generated by a mixture of multiple Gaussian distributions (also called components). Each classification corresponds to one of these Gaussian components, and each point in the data set belongs to a classification with a certain probability. It provides soft assignment, which means that each data point has a probability of belonging to each classification.

[0072] The feature vector f of each segment i The probability of is given by the sum of the probabilities of it belonging to each Gaussian component:

[0073]

[0074] in, is the multivariate Gaussian (normal) distribution of cluster j:

[0075]

[0076] Where: μ j is the mean of the jth Gaussian component. j is the covariance matrix of the jth Gaussian component. d is the dimension (number of features) of the eigenvector. |∑ j | is the determinant of the covariance matrix, and is its inverse. j is the mixing coefficient of the jth Gaussian, representing the proportion of data points that belong to this component. These must sum to 1:

[0077]

[0078] Since the present invention uses unsupervised learning to perform classification, the data of 119 hydraulic supports of the 81308 working face of a certain mine is taken as an example for explanation.

[0079] The present invention uses unsupervised learning to classify. However, before this invention, coal cutting cycle analysis was conducted, and data of 119 hydraulic supports in 81,308 working faces for 3 months, 667 coal cutting cycles, and 2,001 feature segments were manually sorted, and a category label was set for each segment. Therefore, for the present invention, the accuracy of the method can be verified using data with existing labels. At the same time, the present invention also provides a method for how to verify accuracy in the case of unlabeled data as a reference.

[0080] Accuracy verification method with labeled data:

[0081] Accuracy calculation method: Compare the actual label of each sample with the label predicted by the model. If the two are the same, the prediction is correct. The calculation formula is as follows:

[0082]

[0083] in:

[0084] n represents the number of correctly predicted samples, and N represents the total number of samples.

[0085] TP, TN, and TC represent true examples in the three categories, respectively, that is, the model correctly predicts the samples belonging to a certain category.

[0086] FP, FN, and FC represent errors in three categories, respectively, which means that the model incorrectly predicts samples that do not belong to a certain category.

[0087] Through calculation, the overall classification accuracy is 86.6%, and the overall correctly classified samples are 1734. In the initial support stage, 618 samples are classified correctly, with a single category accuracy of 92.6%, in the working stage, 585 samples are classified correctly, with a single category accuracy of 87.7%, and in the unloading stage, 531 samples are classified correctly, with a single category accuracy of 79.6%.

[0088] From the results, we can see that the recognition effect of the initial support stage is the best, because the characteristics of the initial support stage are simple, obvious and easy to identify. The second is the working stage, which has a better recognition effect, because the data of the working stage is relatively flat. Finally, the recognition of the unloading stage has a relatively poor effect, because the unloading stage has complex characteristics such as rising first and then falling.

[0089] The accuracy verification method for unlabeled data is as follows:

[0090] The distance method is used for evaluation, and the distance is used as a measure of similarity. The smaller the distance between two segments, the more similar they are considered.

[0091] Set the distance threshold to [0,2]. This threshold is an empirical value. The mode or average of the distances between two fragments in the classification can be used as the upper limit. If it is within the threshold range, it means that the classification is correct.

[0092] Four smoothed time series segments of the unloading phase are selected to briefly illustrate the accuracy assessment method.

[0093] Segment1:[27.42, 27.46, 27.5, 27.7, 27.9, 28.1, 28.3, 28.5, 27.42, 26.34, 25.26, 24.18, 13.1]

[0094] Segment2:[34.5, 34.3, 34.48, 34.66, 34.84, 35.02, 35.2, 35.12, 35.04, 34.96, 34.88, 34.8, 32.94, 31.08, 29.22, 27.36, 5.5]

[0095] Segment3:[28.5, 28.5, 28.5, 28.5, 28.5, 27.18, 25.86, 24.54, 23.22, 11.9]

[0096] Segment4:[27.18, 26.94, 26.7, 26.86, 27.02, 27.18, 27.34, 27.5, 25.8, 24.1, 22.4, 20.7, 9]

[0097] The calculation results of each group feature vector are shown in Table 1.

[0098] Table 1

[0099] Segment Mean Std Entropy Segment1 26.09 3.8 3.68 Segment2 31.99 6.9 4.03 Segment3 25.52 4.7 3.29 Segment4 24.51 4.8 3.66

[0100] The distance calculation between fragments is shown in Table 2.

[0101] Table 2

[0102] Segment1 Segment2 Segment3 Segment4 Segment1 0 6.6 1.18 1.8 Segment2 6.6 0 6.8 7.7 Segment3 1.18 6.8 0 1.0 Segment4 1.8 7.7 1.0 0

[0103] As can be seen from the table, the distances between segment 1 and segments 3 and 4 are relatively small (1.18 and 1.8, respectively), which means that these segments have high similarity in terms of feature values ​​(mean, standard deviation, and entropy).

[0104] In contrast, the distance between segment 2 and segment 1 is much larger (6.6), which indicates that segment 2 is significantly different from segment 1. Therefore, for the four groups of data in this example, Segment 2 is obviously far away from the other three groups of data and is not within the threshold range, so it can be understood that the number of correct classifications is 3 and the accuracy is 75%.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for classifying support motion characteristics based on hydraulic support pressure data of fully mechanized mining working face, characterized in that: The method comprises the following steps: Perform wavelet denoising on the pressure data of hydraulic supports to smooth the original signal; Using change point detection technology to segment the smoothed pressure data into different stages, the stages include an initial support stage, a working stage, and an unloading stage; Extracting features for each stage, the features including mean, standard deviation and entropy; Using unsupervised learning methods, each phase is classified into the initial support phase, the working phase, or the unloading phase according to the characteristics.

2. A method for classifying support motion characteristics based on hydraulic support pressure data of fully mechanized mining working face according to claim 1, characterized in that: The unsupervised learning method is a Gaussian mixture model.

3. The support motion feature classification method based on the hydraulic support pressure data of the fully mechanized mining working face according to claim 1 is characterized by: The change point detection technique is the Pruned Exact Linear Time (PELT) algorithm.

4. The method for classifying support motion characteristics based on hydraulic support pressure data of fully mechanized mining working face according to claim 1 is characterized in that: The wavelet denoising comprises the following steps: Perform continuous wavelet transform on time series data, convert it into wavelet domain, and obtain wavelet coefficients; In the wavelet domain, threshold processing is applied to each coefficient, and the wavelet coefficients with absolute values ​​less than the threshold are set to zero; The signal is reconstructed by inverse wavelet transform using the wavelet coefficients after threshold processing to obtain the denoised signal.

5. The method for classifying support motion characteristics based on hydraulic support pressure data of fully mechanized mining working face according to claim 1 is characterized in that: The feature extraction for each stage comprises the following steps: Calculate the average value for each stage; Calculate the standard deviation for each stage; Calculate the entropy at each stage.

6. The method for classifying support motion characteristics based on hydraulic support pressure data of fully mechanized mining working face according to claim 1 is characterized in that: The classification of each stage into the initial support stage, the working stage or the unloading stage according to the characteristics specifically comprises the following steps: Perform probability calculation on the feature vector of each stage to obtain the probability that each stage belongs to each Gaussian component; Based on the probability calculation results, each stage is classified as the initial support stage, working stage or unloading stage.

7. The method for classifying support motion characteristics based on hydraulic support pressure data of fully mechanized mining working face according to claim 2 is characterized in that: In the Gaussian mixture model, each Gaussian component corresponds to a stage of the support movement.

Images