Coal mining industry multi-dimensional data real-time monitoring and anomaly positioning method and system

Abstract

The invention discloses a coal mine industry multi-dimensional data real-time monitoring and abnormity positioning method and system, and the method comprises the steps: collecting real-time data in coal mine operation, and carrying out the preprocessing; and carrying out dimension construction on the preprocessed data, and fusing the data of different dimensions to form multi-dimensional data. And inputting the multi-dimensional data into the anomaly detection model, and carrying out backtracking analysis on the abnormal data. And performing exception handling and response according to the detected exception and backtracking analysis. According to the coal mining industry multi-dimensional data real-time monitoring and anomaly positioning method and system provided by the invention, correlation analysis and fusion are performed on different-dimensional data, key features are extracted by using principal component analysis, and efficient fusion of the multi-dimensional data is ensured in combination with nonlinear transformation. The main information is reserved, and the accuracy of anomaly detection is improved by enhancing the representativeness of the features. The abnormal response time is shortened, and the emergency processing accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of data monitoring technology, and in particular to a method and system for real-time monitoring and abnormality positioning of multi-dimensional data in the coal mining industry. Background Art

[0002] With the continuous development of coal mining technology, data collection and monitoring systems in coal mine operating environments have become an important means to ensure safe production. In the existing technology, coal mine monitoring systems are mostly based on single-dimensional or limited-dimensional data collection and monitoring, such as monitoring of environmental parameters (such as temperature, humidity and gas concentration), or single monitoring of equipment status. Although these systems can improve the safety of coal mine operations to a certain extent, due to the limitations of data collection and analysis dimensions, they often cannot fully reflect the overall situation of coal mine operations. Especially in the face of complex abnormal situations, the accuracy and response speed of existing monitoring systems are relatively limited.

[0003] The existing coal mine monitoring system still has some shortcomings. First, the problem of isolated data processing is more prominent, and the correlation between data of different dimensions has not been fully explored, resulting in the system's limited ability to perceive abnormal situations. Secondly, many systems can only detect anomalies, but lack the precise location of the source of the anomaly, and cannot provide strong decision-making support for subsequent emergency response. In addition, the existing system lacks flexibility and accuracy in data preprocessing. When faced with the fusion and analysis of multi-dimensional data, there are often problems of data redundancy or information loss. These defects directly affect the monitoring effect of coal mine safety production. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the problem of isolated data processing in the prior art is more prominent, and the correlation between data of different dimensions has not been fully explored, resulting in the system's limited ability to perceive abnormal situations. Many systems can only detect anomalies, but lack the precise location of the source of the anomaly, and cannot provide strong decision-making support for subsequent emergency response. In addition, the existing system lacks flexibility and accuracy in data preprocessing, and when faced with the fusion and analysis of multi-dimensional data, there are often problems of data redundancy or information loss. These defects directly affect the monitoring effect of coal mine safety production.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: a method for real-time monitoring and abnormality positioning of multi-dimensional data in coal mining, comprising: collecting real-time data in coal mining operations and performing preprocessing.

[0007] The preprocessed data is dimensioned and data of different dimensions are fused to form multidimensional data.

[0008] The multi-dimensional data is input into the anomaly detection model to perform retrospective analysis on the anomaly data.

[0009] Perform exception handling and response based on detected anomalies and backtracking analysis.

[0010] As a preferred solution of the method for real-time monitoring and anomaly positioning of multi-dimensional data in the coal mining industry described in the present invention, the real-time data collected in coal mining operations and preprocessed include collecting environmental data, equipment data, and operation data.

[0011] Environmental data includes temperature, humidity, and gas concentration; equipment data includes vibration frequency, power, and speed; and operation data includes operation time and operation mode.

[0012] Preprocessing includes missing value processing, outlier processing, and data standardization.

[0013] For numerical data, the average value of the data before and after the missing part is used to fill the missing part. For time series data, linear interpolation is used to fill the missing part.

[0014] Outlier processing uses the Z score method to calculate the Z score of each data point, expressed as:

[0015]

[0016] Among them, x represents the current data point, μ represents the data mean, and σ represents the standard deviation.

[0017] When the Z score exceeds the abnormal threshold, it is considered as abnormal data and discarded.

[0018] Data normalization is expressed as:

[0019]

[0020] Among them, x i ′ represents the standardized data, x i represents the original data, x min Represents the minimum value in the data, x max Indicates the maximum value in the data.

[0021] As a preferred solution of the method for real-time monitoring and anomaly positioning of multi-dimensional data in the coal mining industry described in the present invention, the dimensional construction of the preprocessed data includes classifying the data into three dimensions, including equipment dimension, environment dimension, and operation dimension.

[0022] The equipment dimension includes the equipment status dimension and the equipment performance dimension.

[0023] The equipment status dimension describes the operating status of coal mine equipment, including vibration frequency, power, and speed, expressed as:

[0024] D status =[f,P,N]

[0025] Among them, D status represents the device state vector, f represents the device vibration frequency, P represents the device power, and N represents the device rotation speed.

[0026] The equipment performance dimension describes the performance indicators of the equipment, including efficiency and failure rate, expressed as:

[0027] D performance =[η,λ]

[0028] Among them, D performance represents the equipment performance vector, η represents the equipment efficiency, and λ represents the failure rate.

[0029] The environmental dimension includes the environmental status dimension, which describes the environmental data of the coal mine operation area, including temperature, humidity and gas concentration, expressed as:

[0030]

[0031] Among them, E status represents the environmental state vector, T represents temperature, H represents humidity, represents the methane concentration, Indicates the carbon dioxide concentration.

[0032] The operation dimension includes the operation behavior dimension, which describes the operation behavior data during the coal mine operation process, including operation time and operation mode, expressed as:

[0033] O behavior =[t op ,M op ]

[0034] Among them, O behavior represents the operation behavior vector, t op Indicates the operation time, M op Indicates the operation mode.

[0035] As a preferred solution of the method for real-time monitoring and abnormal location of multi-dimensional data in coal mining industry described in the present invention, the fusion of data of different dimensions to form multi-dimensional data includes, in order to understand the correlation between different dimensions, performing correlation analysis on vectors of different dimensions and calculating the correlation matrix between the dimensions, which is expressed as:

[0036] R=[r ij ]

[0037]

[0038] Among them, R represents the correlation matrix, which describes the linear correlation between data of different dimensions. ij represents the correlation coefficient between the i-th dimension and the j-th dimension, x ik represents the value of dimension i at the kth data point, x jk represents the value of dimension j at the kth data point, μ i represents the mean of dimension i, μ j represents the mean of dimension j.

[0039] The main features are extracted from the data of different dimensions through principal component analysis. The main feature vectors are extracted from the three dimensions of equipment status, environmental status and operation behavior.

[0040] Calculate the covariance matrix of the data, expressed as:

[0041]

[0042] Where C represents the covariance matrix of the data, X k represents the data vector, μ represents the mean vector, Represents the transposed matrix.

[0043] Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and corresponding eigenvectors, and select the eigenvector corresponding to the maximum eigenvalue as the main feature.

[0044] Construct the principal component vector P and use the main eigenvector as the basis for fusing data.

[0045] Before fusing the data, in order to enhance the representativeness of the features, the extracted principal components are subjected to nonlinear transformation, and the Sigmoid transformation is expressed as:

[0046]

[0047] Among them, y represents the transformed principal component, P i represents the i-th extracted principal component.

[0048] Finally, the principal component vectors extracted from each dimension are fused through linear combination and nonlinear activation to generate multidimensional data, which can be expressed as:

[0049] F = Relu(W1P device +W2P environment +W3P operation )

[0050] Among them, F represents the fusion feature vector, W1 represents the weight of the device state principal component vector, P device represents the principal component vector of the equipment state, W2 represents the weight of the principal component vector of the environment state, P environmentrepresents the principal component vector of the environmental state, W3 represents the weight of the principal component vector of the operation behavior, and P operation Represents the principal component vector of the operation behavior, and Relu represents the activation function.

[0051] As a preferred solution of the method for real-time monitoring and anomaly positioning of multi-dimensional data in the coal mining industry described in the present invention, the inputting of multi-dimensional data into an anomaly detection model includes, the anomaly detection model includes an encoder, a decoder, a classifier and an attention retrospective analysis module.

[0052] The encoder and decoder are used to compress and reconstruct the data, the classifier performs anomaly detection, and the attention mechanism is used for retrospective analysis.

[0053] The encoder part receives the input multi-dimensional data vector F, compresses the high-dimensional data into a low-dimensional latent space through a layer-by-layer network, and captures the main features of the data. The output of the encoder is the latent variable z, which contains the key information of the input data.

[0054] The input layer takes the multidimensional data vector F as input, and the fully connected layer 1 performs linear transformation and nonlinear activation on the input data. The activation function uses ReLU, and the output is a higher-dimensional feature vector, which is expressed as:

[0055] h1=ReLU(W1F+b1)

[0056] The fully connected layer 2 further compresses the feature vector and extracts a more compact feature representation, which is expressed as:

[0057] h2=ReLU(W2h1+b2)

[0058] The latent space layer maps the feature vector to a low-dimensional latent space to obtain the latent variable z, which is expressed as:

[0059] z=W z h2+b z

[0060] Among them, h1 represents the fully connected layer 1, h2 represents the fully connected layer 2, W1 represents the weight of the fully connected layer 1, b1 represents the bias of the fully connected layer 1, W2 represents the weight of the fully connected layer 2, b2 represents the bias of the fully connected layer 2, and W z represents the weight of the latent space layer, b z represents the latent space layer bias.

[0061] The decoder part reconstructs the original multidimensional data from the latent space and compares it with the input data F. The support vector machine is used as the classifier, and anomaly detection is performed based on the latent variable z. The latent variable z is input into the support vector machine for binary classification, and the output result is abnormal or non-abnormal, which is expressed as:

[0062] y'=SVM(z)

[0063] Among them, y' represents the classification result, 0 represents non-abnormal, and 1 represents abnormal.

[0064] As a preferred solution of the method for real-time monitoring and anomaly positioning of multi-dimensional data in the coal mining industry described in the present invention, the retrospective analysis of abnormal data includes further analyzing the data with abnormal output results, introducing an attention mechanism to identify which part of the features contributes the most, thereby determining which principal component vector has an abnormality.

[0065] The attention mechanism measures the contribution of each feature to the anomaly by calculating the weight of the input feature. The attention weight formula is expressed as:

[0066]

[0067] Among them, α i represents the attention weight of the i-th feature, W a represents the attention weight matrix, h i Indicates feature representation.

[0068] Identify which principal component vector h through the attention mechanism max The largest contribution to the anomaly is traced back to the equipment dimension, environment dimension, and operation dimension based on the principal component vector.

[0069] Determine which dimension of data causes the anomaly, and find the specific abnormal data point by further analyzing the data of that dimension.

[0070] As a preferred solution of the method for real-time monitoring and anomaly positioning of multi-dimensional data in the coal mining industry described in the present invention, the anomaly processing and response based on the detected anomaly and backtracking analysis includes evaluating the risks caused by the anomaly according to the dimension to which the abnormal data point belongs, taking corresponding response measures according to the risk assessment results, and continuously monitoring the system after taking emergency measures to ensure that the abnormal situation is effectively controlled, recording and analyzing the processed anomalies, and providing a reference for subsequent system optimization and preventive measures.

[0071] A multi-dimensional data real-time monitoring and abnormality positioning system for coal mining, characterized by: comprising:

[0072] The data collection module collects real-time data from coal mine operations and performs pre-processing.

[0073] The fusion module constructs the dimensions of the preprocessed data and fuses data of different dimensions to form multi-dimensional data.

[0074] The computing module inputs the multi-dimensional data into the anomaly detection model and performs retrospective analysis on the anomaly data.

[0075] The response module performs exception processing and response based on the detected exceptions and backtracking analysis.

[0076] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0077] A computer-readable storage medium stores a computer program, which implements the steps of the method described above when executed by a processor.

[0078] The beneficial effects of the present invention are as follows: correlation analysis and fusion of data of different dimensions are performed, key features are extracted using principal component analysis, and nonlinear transformation is combined to ensure efficient fusion of multidimensional data. Not only the main information is retained, but also the accuracy of anomaly detection is improved by enhancing the representativeness of features. The abnormal response time is shortened and the accuracy of emergency handling is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0080] Figure 1 An overall flow chart of a method for real-time monitoring and anomaly location of multi-dimensional data in the coal mining industry provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0081] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0082] Example 1, reference Figure 1 , as an embodiment of the present invention, provides a method, comprising:

[0083] S1: Collect real-time data from coal mine operations and perform preprocessing.

[0084] Collect environmental data, equipment data, and operation data.

[0085] Environmental data includes temperature, humidity, and gas concentration; equipment data includes vibration frequency, power, and speed; and operation data includes operation time and operation mode.

[0086] Preprocessing includes missing value processing, outlier processing, and data standardization.

[0087] For numerical data, the average value of the data before and after the missing part is used to fill the missing part. For time series data, linear interpolation is used to fill the missing part.

[0088] Outlier processing uses the Z score method to calculate the Z score of each data point, expressed as:

[0089]

[0090] Among them, x represents the current data point, μ represents the data mean, and σ represents the standard deviation.

[0091] When the Z score exceeds the abnormal threshold, it is considered as abnormal data and discarded.

[0092] Data normalization is expressed as:

[0093]

[0094] Among them, x i ′ represents the standardized data, x i represents the original data, x min Represents the minimum value in the data, x max Indicates the maximum value in the data.

[0095] S2: Dimension the preprocessed data and fuse data of different dimensions to form multidimensional data.

[0096] The data is classified into three dimensions, including device dimension, environment dimension, and operation dimension.

[0097] The equipment dimension includes the equipment status dimension and the equipment performance dimension.

[0098] The equipment status dimension describes the operating status of coal mine equipment, including vibration frequency, power, and speed, expressed as:

[0099] D status =[f,P,N]

[0100] Among them, D status represents the device state vector, f represents the device vibration frequency, P represents the device power, and N represents the device rotation speed.

[0101] The equipment performance dimension describes the performance indicators of the equipment, including efficiency and failure rate, expressed as:

[0102] Dperformance =[η,λ]

[0103] Among them, D performance represents the equipment performance vector, η represents the equipment efficiency, and λ represents the failure rate.

[0104] The environmental dimension includes the environmental status dimension, which describes the environmental data of the coal mine operation area, including temperature, humidity and gas concentration, expressed as:

[0105]

[0106] Among them, E status represents the environmental state vector, T represents temperature, H represents humidity, represents the methane concentration, Indicates the carbon dioxide concentration.

[0107] The operation dimension includes the operation behavior dimension, which describes the operation behavior data during the coal mine operation process, including operation time and operation mode, expressed as:

[0108] O behavior =[t op ,M op ]

[0109] Among them, O behavior represents the operation behavior vector, t op Indicates the operation time, M op Indicates the operation mode.

[0110] In order to understand the correlation between different dimensions, correlation analysis is performed on vectors of different dimensions, and the correlation matrix between dimensions is calculated, which is expressed as:

[0111] R=[r ij ]

[0112]

[0113] Among them, R represents the correlation matrix, which describes the linear correlation between data of different dimensions. ij represents the correlation coefficient between the i-th dimension and the j-th dimension, x ik represents the value of dimension i at the kth data point, x jk represents the value of dimension j at the kth data point, μ i represents the mean of dimension i, μ j represents the mean of dimension j.

[0114] It should be noted that correlation analysis and principal component analysis (PCA) are used to extract the main features in multidimensional data and reduce data redundancy. Correlation analysis identifies highly correlated variables by calculating the correlation between dimensions, thereby guiding the subsequent data fusion process. Principal component analysis extracts the principal components of the data and transforms multidimensional data into key features in a low-dimensional space, thereby reducing the complexity of the data.

[0115] The main features are extracted from the data of different dimensions through principal component analysis. The main feature vectors are extracted from the three dimensions of equipment status, environmental status and operation behavior.

[0116] Calculate the covariance matrix of the data, expressed as:

[0117]

[0118] Where C represents the covariance matrix of the data, X k represents the data vector, μ represents the mean vector, Represents the transposed matrix.

[0119] Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and corresponding eigenvectors, and select the eigenvector corresponding to the maximum eigenvalue as the main feature.

[0120] Construct the principal component vector P and use the main eigenvector as the basis for fusing data.

[0121] Before fusing the data, in order to enhance the representativeness of the features, the extracted principal components are subjected to nonlinear transformation, and the Sigmoid transformation is expressed as:

[0122]

[0123] Among them, y represents the transformed principal component, P i represents the i-th extracted principal component.

[0124] It should be noted that before multidimensional data fusion, the extracted principal components are subjected to nonlinear transformation, such as Sigmoid or ReLU transformation. This design idea draws on the nonlinear activation function in neural networks and aims to enhance the expressiveness of data features through nonlinear transformation, especially when the data distribution is nonlinear and complex, so as to better capture the nonlinear relationship between data.

[0125] Nonlinear transformation can improve the distinguishability of data features and avoid the information loss that may be caused by simple linear combinations. By performing nonlinear processing on data, the system can better identify complex abnormal patterns and enhance the performance of anomaly detection models, especially when facing multidimensional data with nonlinear relationships, which can improve the accuracy and robustness of the model.

[0126] Finally, the principal component vectors extracted from each dimension are fused through linear combination and nonlinear activation to generate multidimensional data, which can be expressed as:

[0127] F = Relu(W1P device +W2P environment +W3P operation )

[0128] Among them, F represents the fusion feature vector, W1 represents the weight of the device state principal component vector, P device represents the principal component vector of the equipment state, W2 represents the weight of the principal component vector of the environment state, P environment represents the principal component vector of the environmental state, W3 represents the weight of the principal component vector of the operation behavior, and P operation represents the principal component vector of the operation behavior, and Relu represents the activation function.

[0129] S3: Input the multidimensional data into the anomaly detection model and perform retrospective analysis on the anomaly data.

[0130] The anomaly detection model consists of an encoder, a decoder, a classifier, and an attention retrospective analysis module.

[0131] The encoder and decoder are used to compress and reconstruct the data, the classifier performs anomaly detection, and the attention mechanism is used for retrospective analysis.

[0132] The encoder part receives the input multi-dimensional data vector F, compresses the high-dimensional data into a low-dimensional latent space through a layer-by-layer network, and captures the main features of the data. The output of the encoder is the latent variable z, which contains the key information of the input data.

[0133] The input layer takes the multidimensional data vector F as input, and the fully connected layer 1 performs linear transformation and nonlinear activation on the input data. The activation function uses ReLU, and the output is a higher-dimensional feature vector, which is expressed as:

[0134] h1=ReLU(W1F+b1)

[0135] The fully connected layer 2 further compresses the feature vector and extracts a more compact feature representation, which is expressed as:

[0136] h2=ReLU(W2h1+b2)

[0137] The latent space layer maps the feature vector to a low-dimensional latent space to obtain the latent variable z, which is expressed as:

[0138] z=W z h2+b z

[0139] Among them, h1 represents the fully connected layer 1, h2 represents the fully connected layer 2, W1 represents the weight of the fully connected layer 1, b1 represents the bias of the fully connected layer 1, W2 represents the weight of the fully connected layer 2, b2 represents the bias of the fully connected layer 2, and W z represents the weight of the latent space layer, b z represents the latent space layer bias.

[0140] The decoder part reconstructs the original multidimensional data from the latent space and compares it with the input data F. The support vector machine is used as the classifier, and anomaly detection is performed based on the latent variable z. The latent variable z is input into the support vector machine for binary classification, and the output result is abnormal or non-abnormal, which is expressed as:

[0141] y'=SVM(z)

[0142] Among them, y' represents the classification result, 0 represents non-abnormal, and 1 represents abnormal.

[0143] S4: Perform exception handling and response based on detected anomalies and backtracking analysis.

[0144] The data with abnormal output results are further analyzed, and the attention mechanism is introduced to identify which part of the features contributes the most, so as to determine which principal component vector is abnormal.

[0145] The attention mechanism measures the contribution of each feature to the anomaly by calculating the weight of the input feature. The attention weight formula is expressed as:

[0146]

[0147] Among them, α i represents the attention weight of the i-th feature, W a represents the attention weight matrix, h i Indicates feature representation.

[0148] Identify which principal component vector h through the attention mechanism max The largest contribution to the anomaly is traced back to the equipment dimension, environment dimension, and operation dimension based on the principal component vector.

[0149] Determine which dimension of data causes the anomaly, and find the specific abnormal data point by further analyzing the data of that dimension.

[0150] It should be noted that a method based on gradient backtracking and local explanation model is proposed to analyze the data points that cause anomalies in detail. This process combines the gradient backtracking mechanism in deep learning and uses local linear models to explain abnormal data in complex models.

[0151] Gradient backtracking is used to calculate the sensitivity of input data to abnormal outputs and identify abnormal data points by backtracking gradient information. The Local Interpretation Model (LIME) builds local linear models near abnormal data points to explain abnormal behavior in complex models. This method combines gradient backtracking and model interpretation techniques in deep learning to provide more detailed analysis and explanations for complex anomalies.

[0152] Gradient backtracking can help the system locate abnormal data points more precisely and determine which inputs have the greatest impact on abnormal outputs. The LIME local interpretation model can help the system generate interpretable linear approximations in complex models, making the source and cause of anomalies more transparent. In this way, the system can not only detect anomalies, but also provide interpretable anomaly analysis results, thereby improving the reliability of the system and user trust.

[0153] Calculate the sensitivity of the input data (i.e., each data point in the device dimension, environment dimension, and operation dimension) to abnormal output, that is, the degree of influence of the change of the data point on the output result. This is achieved by calculating the gradient of the model output to the input data.

[0154] Assume that an anomaly has been detected, that is, the SVM output is classified as an anomaly y = 1. Now we need to backtrack and identify the data point that caused this anomaly by calculating the gradient of the model input. Set the relationship between the model output y' and the input vector F to y' = f(F), and calculate the partial derivative (gradient) of the model output for each input component Fj:

[0155]

[0156] in, Represents the model output to input component F j The sensitivity of the input component F j The degree of influence on the output results.

[0157] By calculating each input component F j The gradient of is used to determine which input data point has the greatest impact on the anomaly. The larger the gradient value, the greater the contribution of the data point to the anomaly.

[0158] Furthermore, to further strengthen the analysis, the cumulative gradient of the input data points to the abnormal output of the model is calculated, that is, in different network layers, how the change of the input data affects the final output through each layer. This cumulative gradient can be achieved by back propagating the accumulated gradient values ​​in each layer. The cumulative gradient formula is expressed as:

[0159]

[0160] Among them, G j Represents data point F jThe cumulative gradient of represents its overall impact on the anomaly. L represents the number of layers in the model. Represents the activation value of the jth data point in the lth layer of the model.

[0161] By calculating the cumulative gradient, we can more finely identify which data points contribute most to the anomaly.

[0162] It should be noted that gradient backtracking helps determine which data points have the greatest impact on abnormal outputs, but this is not enough to explain how specific data points in complex models cause anomalies. Therefore, it is necessary to introduce LIME, a method to explain the output of complex models through local linear models. The core idea of ​​LIME is to build a simple linear model near the abnormal data points to explain the behavior of the complex model. The specific steps are as follows:

[0163] For the detected abnormal data points, a set of local data points are generated near them. These data points are generated by applying small perturbations to the original data points, thus forming a local data set {F i ′}.

[0164] The perturbation formula is expressed as:

[0165] F i ′=F i +∈ i

[0166] Among them, F i ′ represents a local data point, ∈ i represents the applied disturbance.

[0167] On the local data set, a simple linear model is constructed to approximate the behavior of the complex model. It can be expressed as:

[0168]

[0169] Among them, w represents the weight vector of the local new model, represents the linear model output, and b represents the bias of the local new model.

[0170] Linear models are fitted by minimizing the weighted squared error of local data points:

[0171]

[0172] Among them, w i Represents the weight of the local data point, which is inversely proportional to the distance to the original data point.

[0173] Furthermore, we combine gradient backtracking with LIME to more accurately identify abnormal data points and explain their abnormal behavior. First, we calculate the sensitivity of each data point to the anomaly through gradient backtracking and identify the groups of data points that contribute most to the anomaly. We generate local data sets near the abnormal data points, build local linear models, and analyze which data components have a significant impact on the anomaly. We combine the sensitivity analysis of gradient backtracking and the local linear model results of LIME to determine the final abnormal data points.

[0174] According to the dimension to which the abnormal data point belongs, the risk caused by the anomaly is assessed, and corresponding response measures are taken according to the risk assessment results. After taking emergency measures, the system is continuously monitored to ensure that the abnormal situation is effectively controlled, and the processed anomalies are recorded and analyzed to provide a reference for subsequent system optimization and preventive measures.

[0175] The above embodiments also include a multi-dimensional data real-time monitoring and abnormality positioning system for coal mining, specifically:

[0176] The data collection module collects real-time data from coal mine operations and performs pre-processing.

[0177] The fusion module constructs the dimensions of the preprocessed data and fuses data of different dimensions to form multi-dimensional data.

[0178] The computing module inputs the multi-dimensional data into the anomaly detection model and performs retrospective analysis on the anomaly data.

[0179] The response module performs exception processing and response based on the detected exceptions and backtracking analysis.

[0180] The computer device may be a server. The computer device includes a processor, a memory, an input / output interface (I / O for short) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data cluster data of a power monitoring system. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a method is implemented.

[0181] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0182] Example 2, an embodiment of the present invention, provides a method and system for real-time monitoring of multi-dimensional data and abnormality positioning in the coal mining industry. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0183] The specific steps of our invention are:

[0184] By installing sensors, the equipment operation status, environmental status and operation behavior data are collected and preprocessed. The preprocessed data is divided into equipment dimension, environmental dimension and operation dimension, correlation analysis is performed, the principal components are extracted and the feature expression is enhanced through nonlinear transformation. The multi-dimensional data is fused into a multi-dimensional state vector. The multi-dimensional state vector is input into the anomaly detection model, and the variational autoencoder (VAE) and support vector machine (SVM) are used for anomaly detection. The specific abnormal data points are located through the attention mechanism and gradient backtracking method. According to the detected anomalies, combined with the backtracking analysis results, the risk assessment of the abnormal points is carried out and corresponding emergency measures are taken.

[0185] The steps of the existing technology are: only monitor the vibration frequency and power of the equipment, and use simple threshold judgment to detect abnormalities. Once it is detected that the equipment parameters exceed the threshold, an alarm is automatically triggered and the equipment is shut down for inspection, but the specific cause is not analyzed. The experimental results are shown in Table 1.

[0186] Table 1

[0187]

[0188]

[0189] Through experimental data comparison, our invention has shown significant advantages in multi-dimensional data monitoring, anomaly detection and retrospective analysis. The existing technology only monitors the vibration frequency and power in the equipment dimension. In the data of the fifth day, when the equipment vibration frequency reached 25Hz and the power reached 190kW, the system triggered an abnormal alarm, but the system failed to identify the specific source of the abnormality, resulting in a long shutdown inspection.

[0190] In contrast, our invention detects potential risks in advance in multiple indicators of the device dimension, environmental dimension, and operation dimension through multi-dimensional data fusion. On the fifth day, although the equipment status score was only 0.07, combined with the abnormal detection results of the environmental dimension (score 0.08) and the operation dimension (score 0.09), the system comprehensively assessed high-risk risks (score 0.24). Based on retrospective analysis, our invention system can accurately locate the abnormality of the operation mode and avoid more serious failures through early intervention.

[0191] This experiment shows that the multi-dimensional data monitoring and anomaly location method invented by us can not only detect potential risks earlier and more accurately, but also provide specific abnormal causes through retrospective analysis, reduce unnecessary downtime and inspection time, and thus significantly improve the safety and operational efficiency of coal mine operations.

[0192] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for real-time monitoring and abnormal location of multi-dimensional data in coal mining, characterized in that: include: Collect real-time data from coal mine operations and perform pre-processing; Dimensionalize the preprocessed data and fuse data of different dimensions to form multidimensional data; Input multidimensional data into the anomaly detection model and perform retrospective analysis on the anomaly data; Perform exception handling and response based on detected anomalies and backtrace analysis.

2. The method for real-time monitoring and abnormality location of multi-dimensional data in the coal mining industry according to claim 1, characterized in that: The real-time data collected in coal mine operations and preprocessed include collecting environmental data, equipment data, and operation data; Environmental data includes temperature, humidity, and gas concentration; equipment data includes vibration frequency, power, and speed; and operation data includes operation time and operation mode; Preprocessing includes missing value processing, outlier processing, and data standardization; For numerical data, the average value of the data before and after the missing part is used to fill the missing part. For time series data, linear interpolation is used to fill the missing part. Outlier processing uses the Z score method to calculate the Z score of each data point, expressed as: Among them, x represents the current data point, μ represents the data mean, and σ represents the standard deviation; When the Z score exceeds the abnormal threshold, it is considered as abnormal data and discarded; Data normalization is expressed as: Among them, x i ′ represents the standardized data, x i represents the original data, x min Represents the minimum value in the data, x max Indicates the maximum value in the data.

3. The method for real-time monitoring and abnormality location of multi-dimensional data in the coal mining industry according to claim 2, characterized in that: The dimension construction of the pre-processed data includes classifying the data into three dimensions, including device dimension, environment dimension, and operation dimension; The equipment dimension includes the equipment status dimension and the equipment performance dimension; The equipment status dimension describes the operating status of coal mine equipment, including vibration frequency, power, and speed, expressed as: D status =[f,P,N] Among them, D status represents the device state vector, f represents the device vibration frequency, P represents the device power, and N represents the device speed; The equipment performance dimension describes the performance indicators of the equipment, including efficiency and failure rate, expressed as: D performance =[h,l] Among them, D performance represents the equipment performance vector, η represents the equipment efficiency, and λ represents the failure rate; The environmental dimension includes the environmental status dimension, which describes the environmental data of the coal mine operation area, including temperature, humidity and gas concentration, expressed as: Among them, E status represents the environmental state vector, T represents temperature, H represents humidity, represents the methane concentration, Indicates the concentration of carbon dioxide; The operation dimension includes the operation behavior dimension, which describes the operation behavior data during the coal mine operation process, including operation time and operation mode, expressed as: O behavior =[t op ,M op ] Among them, O behavior represents the operation behavior vector, t op Indicates the operation time, M op Indicates the operation mode.

4. The method for real-time monitoring and abnormality location of multi-dimensional data in the coal mining industry according to claim 3, characterized in that: The fusion of data of different dimensions to form multi-dimensional data includes performing correlation analysis on vectors of different dimensions to understand the correlation between different dimensions and calculating the correlation matrix between the dimensions, which is expressed as: R＝[r ij ] Among them, R represents the correlation matrix, which describes the linear correlation between data of different dimensions; r ij represents the correlation coefficient between the i-th dimension and the j-th dimension, x ik represents the value of dimension i at the kth data point, x jk represents the value of dimension j at the kth data point, μ i represents the mean of dimension i, μ j represents the mean of dimension j; Through principal component analysis, the main features are extracted from data of different dimensions; the main feature vectors are extracted from the three dimensions of equipment status, environmental status and operation behavior; Calculate the covariance matrix of the data, expressed as: Where C represents the covariance matrix of the data, X k represents the data vector, μ represents the mean vector, represents the transposed matrix; Perform eigenvalue decomposition on the covariance matrix C to obtain the eigenvalues ​​and corresponding eigenvectors, and select the eigenvector corresponding to the maximum eigenvalue as the main feature; Construct the principal component vector P and use the main eigenvector as the basis for fusion data; Before fusing the data, in order to enhance the representativeness of the features, the extracted principal components are subjected to nonlinear transformation, and the Sigmoid transformation is expressed as: Among them, y represents the transformed principal component, P i represents the i-th extracted principal component; Finally, the principal component vectors extracted from each dimension are fused through linear combination and nonlinear activation to generate multidimensional data, which can be expressed as: F=Relu(W1P device +W2P environment +W3P operation ) Among them, F represents the fusion feature vector, W1 represents the weight of the device state principal component vector, P device represents the principal component vector of the equipment state, W2 represents the weight of the principal component vector of the environment state, P environment represents the principal component vector of the environmental state, W3 represents the weight of the principal component vector of the operation behavior, and P operation represents the principal component vector of the operation behavior, and Relu represents the activation function.

5. The method for real-time monitoring and abnormality location of multi-dimensional data in the coal mining industry according to claim 4, characterized in that: The inputting the multi-dimensional data into the anomaly detection model includes, the anomaly detection model includes an encoder, a decoder, a classifier and an attention backtracking analysis module; The encoder and decoder are used to compress and reconstruct the data, the classifier is used for anomaly detection, and the attention mechanism is used for retrospective analysis; The encoder receives the input multi-dimensional data vector F and compresses the high-dimensional data into a low-dimensional latent space through a layer-by-layer network to capture the main features of the data. The output of the encoder is the latent variable z, which contains the key information of the input data. The input layer takes the multidimensional data vector F as input, and the fully connected layer 1 performs linear transformation and nonlinear activation on the input data. The activation function uses ReLU, and the output is a higher-dimensional feature vector, which is expressed as: h1=ReLU(W1F+b1) The fully connected layer 2 further compresses the feature vector and extracts a more compact feature representation, which is expressed as: h2=ReLU(W2h1+b2) The latent space layer maps the feature vector to a low-dimensional latent space to obtain the latent variable z, which is expressed as: z=W z h2+b z Among them, h1 represents the fully connected layer 1, h2 represents the fully connected layer 2, W1 represents the weight of the fully connected layer 1, b1 represents the bias of the fully connected layer 1, W2 represents the weight of the fully connected layer 2, b2 represents the bias of the fully connected layer 2, and W z represents the weight of the latent space layer, b z represents the bias of the latent space layer; The decoder part reconstructs the original multidimensional data from the latent space and compares it with the input data F. The support vector machine is used as a classifier, and anomaly detection is performed based on the latent variable z. The latent variable z is input into the support vector machine for binary classification, and the output result is abnormal or non-abnormal, which is expressed as: y'=SVM(z) Among them, y' represents the classification result, 0 represents non-abnormal, and 1 represents abnormal.

6. The method for real-time monitoring and abnormality location of multi-dimensional data in the coal mining industry according to claim 5, characterized in that: The retrospective analysis of abnormal data includes further analyzing the data with abnormal output results, introducing an attention mechanism to identify which part of the features contributes the most, thereby determining which principal component vector is abnormal; The attention mechanism measures the contribution of each feature to the anomaly by calculating the weight of the input feature. The attention weight formula is expressed as: Among them, α i represents the attention weight of the i-th feature, W a represents the attention weight matrix, h i Representation of characteristic representation; Identify which principal component vector h through the attention mechanism max The largest contribution to the anomaly is traced back to the equipment dimension, environment dimension, and operation dimension based on the principal component vector; Determine which dimension of data causes the anomaly, and find the specific abnormal data point by further analyzing the data of that dimension.

7. The method for real-time monitoring and abnormality location of multi-dimensional data in the coal mining industry according to claim 6, characterized in that: The exception handling and response based on the detected anomalies and retrospective analysis includes evaluating the risks caused by the anomalies according to the dimensions to which the abnormal data points belong, taking corresponding response measures according to the risk assessment results, and continuously monitoring the system after taking emergency measures to ensure that the abnormal situation is effectively controlled, recording and analyzing the handled anomalies, and providing a reference for subsequent system optimization and preventive measures.

8. A coal mining multi-dimensional data real-time monitoring and abnormality positioning system using the method according to any one of claims 1 to 7, characterized in that: Data collection module, collects real-time data from coal mine operations and performs preprocessing; The fusion module constructs the dimensions of the preprocessed data and fuses data of different dimensions to form multi-dimensional data; The computing module inputs the multi-dimensional data into the anomaly detection model and performs retrospective analysis on the anomaly data; The response module performs exception processing and response based on the detected anomalies and backtracking analysis.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images