Online abnormity monitoring method and system for linear movement cutting ore pulp sampler

Abstract

The invention discloses an online anomaly monitoring method and system for a linear movement cutting ore pulp sampler, and relates to the technical field of industrial automation, and the method comprises the steps: executing frequency band energy separation and sliding window statistical analysis on a working condition data set of the ore pulp sampler, and obtaining a multi-dimensional feature matrix; performing weight distribution and dynamic weighted aggregation on the multi-dimensional feature matrix to form a space-time analysis data packet, performing collaborative analysis on the space-time analysis data packet, and outputting a trend collaborative interaction matrix; and performing risk quantification and contribution degree distribution on the trend collaborative interaction matrix by using an entropy weight method to generate an abnormal quantification parameter, and performing confidence coefficient weighted calculation on the abnormal quantification parameter to form an abnormal probability value. According to the method, the working condition data set of the ore pulp sampler is fully fused through the sliding window statistical analysis and the entropy weight method, and meanwhile, deep feature mining and spatial relation fusion are performed through the dynamic causal atlas and the space-time convolutional neural network model, so that the reliability of anomaly monitoring is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial automation, and in particular to an online abnormality monitoring method and system for a linearly moving cutting ore pulp sampler. BACKGROUND

[0002] With the continuous improvement of industrial automation and intelligentization, the linearly moving cutting ore pulp sampler, as one of the key equipment in the mining production process, gradually attracts widespread attention in terms of monitoring and maintenance of the running state of the equipment. The traditional monitoring method mainly relies on manual inspection and regular maintenance. In recent years, with the rapid development of sensing technology, data acquisition system and edge computing equipment, intelligent operation and maintenance methods based on online monitoring and data analysis have been gradually applied to the state perception and fault warning of the ore pulp sampler.

[0003] Although the existing monitoring method has made certain progress, there are still some deficiencies. First, in the face of the fusion problem of multi-source heterogeneous data, the traditional monitoring method usually adopts a simple weighted average or direct splicing method to process data, resulting in the loss of important information and the neglect of the relationship between features. Second, the traditional abnormality monitoring system relies on shallow classification models or LSTM sequence models, which are difficult to describe the dynamic behavior of the ore pulp sampler under complex working conditions and the complex causal relationship between components, thereby reducing the reliability of abnormality detection. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides an online abnormality monitoring method for a linearly moving cutting ore pulp sampler to solve the problems of insufficient multi-source data fusion and insufficient dynamic causal relationship modeling.

[0006] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides an online abnormality monitoring method for a linearly moving cutting ore pulp sampler, which includes performing frequency band energy separation and sliding window statistical analysis on the working condition data set of the ore pulp sampler, obtaining a multi-dimensional feature matrix, and performing weight distribution and dynamic weighted aggregation on the multi-dimensional feature matrix to form a space-time analysis data packet; Performing a synergistic analysis on the space-time analysis data packet outputs a trend synergistic interaction matrix; using an entropy weight method to perform risk quantification and contribution degree distribution on the trend synergistic interaction matrix generates an abnormality quantization parameter, and performs confidence weight calculation on the abnormality quantization parameter to form an abnormality probability value; using a time decay compensation factor to perform time sequence correction and spatial correlation mapping on the abnormality probability value obtains a space-time abnormality index vector; The spatio-temporal anomaly index vector is subjected to correlation strength analysis and singular value decomposition, state features and correlation features are extracted, the state features are mapped to nodes of a topology structure, the correlation features are mapped to edges of the topology structure, and a dynamic causal graph is constructed; The dynamic causal graph is input into a spatio-temporal convolutional neural network model, a feature aggregation layer applies an adjacency normalization operator to perform node feature extraction and spatial relationship fusion, a spatio-temporal coupling layer performs multi-scale feature enhancement through a residual cross-layer connection mechanism, an abnormal influence degree score is formed, the abnormal influence degree score is graded and structured, and an abnormal monitoring report is output.

[0007] As a preferred scheme of the linear moving cutting ore pulp sampler online anomaly monitoring method, the ore pulp sampler working condition data set includes mechanical moving cutting parameters, ore pulp physical property parameters, and ore pulp sampler electrical parameters.

[0008] As a preferred scheme of the linear moving cutting ore pulp sampler online anomaly monitoring method, the formation of the spatio-temporal analysis data packet specifically includes the following steps, The ore pulp sampler working condition data set is subjected to frequency band energy separation through fast Fourier transform to form a time-frequency composite parameter sequence, and the time-frequency composite parameter sequence is subjected to sliding window statistical analysis to obtain a multi-dimensional feature matrix; The multi-dimensional feature matrix is subjected to weight distribution and vector reconstruction to generate a weighted spatio-temporal vector, and the weighted spatio-temporal vector is subjected to dynamic weighted aggregation to output the spatio-temporal analysis data packet.

[0009] As a preferred scheme of the linear moving cutting ore pulp sampler online anomaly monitoring method, the acquisition of the spatio-temporal anomaly index vector specifically includes the following steps, The abnormal probability value is subjected to window smoothing and time weight distribution through a time attenuation compensation factor to generate an attenuation correction sequence, and the abnormal probability value is subjected to time sequence correction according to the attenuation correction sequence to generate a time sequence correction vector; The time sequence correction vector is subjected to region division and feature aggregation to obtain a spatial feature set, and the spatial feature set is subjected to spatial correlation mapping to generate the spatio-temporal anomaly index vector.

[0010] As a preferred scheme of the linear moving cutting ore pulp sampler online anomaly monitoring method, the construction of the dynamic causal graph specifically includes the following steps, The spatio-temporal anomaly index vector is subjected to correlation strength analysis to form a correlation strength matrix, and the correlation strength matrix is subjected to singular value decomposition to extract state features and correlation features; The state features are subjected to dimension normalization to obtain standard state parameters, and the standard state parameters are subjected to node attribute mapping to generate nodes of a topology structure. The interaction of the correlation features is quantified to generate an edge weight matrix, the edge weight matrix is subjected to edge strength mapping to obtain edges of the topological structure; Dynamic hierarchical clustering is performed on the nodes of the topological structure and the edges of the topological structure to construct a dynamic causal graph.

[0011] As a preferred scheme of the online anomaly monitoring method of the linear moving cutting ore pulp sampler, wherein: the spatio-temporal convolutional neural network model is constructed by building a feature aggregation layer and a spatio-temporal coupling layer, and applying a multi-head attention mechanism for cross-layer parameterization stacking.

[0012] As a preferred scheme of the online anomaly monitoring method of the linear moving cutting ore pulp sampler, wherein: the formation of the anomaly influence degree score specifically includes the following steps, The dynamic causal graph is input into the spatio-temporal convolutional neural network model, the feature aggregation layer applies an adjacency normalization operator to perform node feature extraction and spatial relationship fusion to generate a primary node embedding representation; The spatio-temporal coupling layer performs multi-scale feature enhancement through a residual cross-layer connection mechanism to obtain an enhanced spatio-temporal feature matrix; The primary node embedding representation and the enhanced spatio-temporal feature matrix are subjected to full connection transformation and probability normalization to form the anomaly influence degree score.

[0013] In a second aspect, the present application provides an online anomaly monitoring system for a linear moving cutting ore pulp sampler, comprising, The data analysis module is used to perform frequency band energy separation and sliding window statistical analysis on the ore pulp sampler working condition data set to obtain a multi-dimensional feature matrix, and perform weight distribution and dynamic weighted aggregation on the multi-dimensional feature matrix to form a spatio-temporal analysis data packet; The anomaly quantification module is used to perform synergy analysis on the spatio-temporal analysis data packet to output a trend synergy interaction matrix; the entropy weight method is used to perform risk quantification and contribution degree distribution on the trend synergy interaction matrix to generate anomaly quantification parameters, and the anomaly quantification parameters are subjected to confidence weighted calculation to form an anomaly probability value; a time decay compensation factor is used to perform time sequence correction and spatial correlation mapping on the anomaly probability value to obtain a spatio-temporal anomaly index vector; The graph construction module is used to perform correlation strength analysis and singular value decomposition on the spatio-temporal anomaly index vector to extract state features and correlation features, and map the state features to nodes of the topological structure and the correlation features to edges of the topological structure to construct a dynamic causal graph; The report generation module is configured to input the dynamic causal graph into a spatio-temporal convolutional neural network model, a feature aggregation layer is configured to apply an adjacency normalization operator to perform node feature extraction and spatial relationship fusion, and a spatio-temporal coupling layer is configured to perform multi-scale feature enhancement through a residual cross-layer connection mechanism to form an abnormal influence degree score, and the abnormal influence degree score is graded and structured to output an abnormal monitoring report.

[0014] In a third aspect, the present application provides a computer device comprising a memory and a processor, and the memory stores a computer program, wherein the computer program is executed by the processor to implement any step of the linear moving cutting ore pulp sampler online abnormal monitoring method according to the first aspect of the present application.

[0015] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program is executed by a processor to implement any step of the linear moving cutting ore pulp sampler online abnormal monitoring method according to the first aspect of the present application.

[0016] The present application has the following beneficial effects: the sliding window statistical analysis and the entropy weight method are used to process the working condition data set of the ore pulp sampler, so that the multi-source data can be more fully fused, and the interaction and potential correlation between different source data can be accurately captured. The dynamic causal graph and the spatio-temporal convolutional neural network model are used for deep feature mining and spatial relationship fusion, which effectively describes the complex causal relationship of the ore pulp sampler under complex working conditions, significantly improves the recognition ability of complex fault modes, and further improves the reliability of the abnormal monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Fig. 1 It is a flowchart of the linear moving cutting ore pulp sampler online abnormal monitoring method.

[0019] Fig. 2 It is a schematic diagram of the linear moving cutting ore pulp sampler online abnormal monitoring system.

[0020] Fig. 3 It is a flowchart of the spatio-temporal abnormal index vector generation.

[0021] Fig. 4 It is a flowchart of the dynamic causal graph construction. DETAILED DESCRIPTION

[0022] In order to make the above objectives, characteristics and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0023] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other ways different from those described herein, and those skilled in the art can make similar generalizations without departing from the connotation of the present application, therefore the present application is not limited by the specific embodiments disclosed below.

[0024] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is separate or alternative to other embodiments.

[0025] Reference Figs. 1-4 For one embodiment of the present application, the embodiment provides an online anomaly monitoring method for a linearly moving cutting ore pulp sampler, comprising the following steps: S1, performing sliding window statistical analysis on the ore pulp sampler working condition data set to obtain a multi-dimensional feature matrix, performing dynamic weighted aggregation on the multi-dimensional feature matrix to form a space-time analysis data package.

[0026] Specifically includes the following operations, S1.1, collecting the ore pulp sampler working condition data set, the ore pulp sampler working condition data set including mechanical moving cutting parameters, ore pulp physical property parameters and ore pulp sampler electrical parameters.

[0027] The mechanical moving cutting parameters include displacement amount, cutting speed, motion acceleration and positioning deviation data; the displacement amount data is collected by a displacement sensor, the cutting speed data is collected by a servo encoder, the motion acceleration data is collected by an acceleration sensor, and the positioning deviation data is collected by a laser range finder; The ore pulp physical property parameters include ore pulp concentration, density, particle size distribution, flow rate and temperature; the ore pulp concentration is collected by an online concentration meter, the density is collected by a mass flow meter, the particle size distribution is collected by a laser particle size analyzer, the flow rate is collected by an ultrasonic flowmeter, and the temperature is collected by a temperature transmitter; The ore pulp sampler electrical parameters include voltage, current, power, frequency and motor winding temperature, the voltage is collected by a voltage sensor, the current is collected by a current transformer, the power is collected by a power analyzer, the frequency is collected by a frequency meter, and the motor winding temperature is collected by a temperature sensor; The ore pulp sampler working condition data set can not only comprehensively reflect the equipment running state and the ore pulp physical property characteristics, but also can improve the accuracy and efficiency of mineral geological exploration services.

[0028] S1.2, pre-process the ore pulp sampler working condition data set, in specific operation, for mechanical movement cutting parameters, apply sliding mean filtering for noise suppression, and use linear interpolation to perform missing value filling, improve the integrity of the mechanical movement cutting parameters, simultaneously through Z-score standardization for dimension scaling, ensure the dimension uniformity; for ore pulp physical property parameters, adopt wavelet transform for multi-layer decomposition and noise removal, to improve the data purity and resolution of ore pulp physical property parameters, simultaneously use sliding window for time series alignment, ensure the synchronization and consistency of ore pulp physical property parameters; for ore pulp sampler electrical parameters, remove high-frequency noise through low-pass filtering, and use sliding window to perform data smoothing, to enhance the continuity of ore pulp sampler electrical parameters, simultaneously use NTP time stamp synchronization protocol for time synchronization, ensure the time consistency of ore pulp sampler electrical parameters.

[0029] S1.3, apply fast Fourier transform to the pre-processed ore pulp sampler working condition data set to perform frequency band energy separation, form time-frequency composite parameter sequence, in specific operation, filter smooth the pre-processed ore pulp sampler working condition data set, and apply ADC converter for digital-analog conversion, obtain digital signal, at the same time, execute Hanning window on the digital signal, generate windowed time domain signal; apply fast Fourier transform to the windowed time domain signal to perform frequency band energy separation, further, perform frequency band decomposition and band-pass filtering on the windowed time domain signal, output complex spectrum containing amplitude and phase information; extract the core frequency band (0-500Hz) of the complex spectrum, and equally divide the core frequency band into 8 sub-bands, perform energy integration on each sub-band to obtain frequency band energy; then perform energy normalization on the frequency band energy to obtain normalized energy distribution, extract frequency domain features of the normalized energy distribution and perform multi-dimensional feature reconstruction to obtain frequency domain feature vector; perform sliding window sampling on the frequency domain feature vector to obtain time-frequency composite parameters, and perform dynamic weight allocation and weighted aggregation on the time-frequency composite parameters to output time-frequency composite parameter sequence.

[0030] S1.4, perform sliding window statistical analysis on the time-frequency composite parameter sequence to obtain a multi-dimensional feature matrix, in specific operation, use a fixed length time window (such as 60 seconds) to slide the time-frequency composite parameter sequence, eliminate abnormal windows with incomplete data, output valid data window; in the valid data window, perform moving exponential average to obtain normalized range, and perform sliding window aggregation on the normalized range to obtain basic statistical parameters; scale and reorganize the basic statistical parameter features to generate multi-scale features, horizontally splice and dimensionally align the multi-scale features to form a multi-dimensional feature matrix.

[0031] S1.5, weight distribution and vector reconstruction are performed on the multi-dimensional feature matrix to generate a weighted space-time vector; the weighted space-time vector is dynamically weighted and aggregated to output a space-time analysis data packet. In the weight distribution stage, linear transformation and weight distribution are performed on the multi-dimensional feature matrix to generate a weighted feature matrix, and the weighted feature matrix is simultaneously subjected to sparse dimension reduction and orthogonalization processing to output a weighted optimized feature set; in the vector reconstruction stage, tensor decomposition is performed on the weighted optimized feature set to extract spatial correlation features, and the spatial correlation features are subjected to feature space projection and three-dimensional tensor reconstruction to form a weighted space-time vector; in the weighted aggregation stage, feature concatenation and weighted aggregation are performed on the weighted space-time vector to obtain primary space-time aggregation data, and the primary space-time aggregation data is simultaneously subjected to time series alignment and data smoothing to output a space-time analysis data packet.

[0032] S2, the entropy weight method is used to perform confidence weighted calculation on the space-time analysis data packet to obtain an abnormal probability value, and a time decay compensation factor is used to perform time series correction and spatial correlation mapping on the abnormal probability value to generate a space-time anomaly index vector.

[0033] Specifically includes the following operations, S2.1, the space-time analysis data packet is subjected to synergistic analysis to output a trend synergistic interaction matrix. In the specific operation, the space-time analysis data packet is uniformly sampled to generate a random feature point set; the K nearest neighbor algorithm is used to perform neighborhood relationship analysis on the random feature point set; further, the random feature point set is subjected to distance measurement and similarity sorting to generate a neighborhood relationship matrix, the neighborhood relationship matrix is subjected to weight distribution and feature aggregation to form a standardized correlation matrix, the cross correlation method is used to perform synergistic analysis on the standardized correlation matrix; further, the standardized correlation matrix is subjected to symmetry processing to obtain an optimized correlation matrix, the optimized correlation matrix is subjected to eigenvalue decomposition to decompose the optimized correlation matrix into eigenvalues and eigenvectors, and principal component extraction is simultaneously performed: the first three principal component features of the eigenvalues and eigenvectors are extracted, and vector normalization and dimension reconstruction are performed to form a synergistic feature vector, the synergistic feature vector is subjected to tensor expansion and matrix conversion to obtain a trend synergistic interaction matrix.

[0034] S2.2, the entropy weight method is used for risk quantization and contribution degree allocation of the trend collaborative interaction matrix, abnormal quantization parameters are generated, confidence weighting calculation is performed on the abnormal quantization parameters, and an abnormal probability value is obtained. In the risk quantization stage, kernel density estimation is used to fit the probability density of the trend collaborative interaction matrix. Further, the trend collaborative interaction matrix is subjected to equal-width discretization to obtain a discretized feature interval. The frequency of the discretized feature interval is counted to obtain a feature frequency distribution. The feature frequency distribution is subjected to probability density normalization to generate a probability distribution matrix. The probability distribution matrix is subjected to logarithmic transformation to output a risk distribution vector. In the contribution degree allocation stage, the risk distribution vector is subjected to weighted summation matrix multiplication to obtain a weighted risk score. According to the weighted risk score, the risk distribution vector is subjected to contribution degree allocation to form a risk contribution degree vector. The risk contribution degree vector is subjected to linear combination and dimension fusion to obtain abnormal quantization parameters; In the weighted calculation stage, weight allocation and parameter aggregation are performed on the abnormal quantization parameters to obtain a weighted parameter matrix. The entropy weight method is applied to the weighted parameter matrix to measure the information entropy. Further, the weighted parameter matrix is subjected to orthogonal decomposition and weighted averaging to obtain an entropy weight initial value. The entropy weight initial value is input into a Shannon entropy formula for confidence weighting calculation to obtain an optimized information entropy value. The Softmax function is applied to the optimized information entropy value for nonlinear transformation to output an abnormal probability value. The specific mathematical formula is as follows, ; Wherein, represents the abnormal probability value, represents a natural constant, represents the optimized information entropy value, represents a standard entropy value. It should be noted that the standard entropy value is defined based on the historical distribution median of the optimized information entropy value.

[0035] S2.3, the time decay compensation factor is used for window smoothing and time weight allocation of the abnormal probability value, a decay correction sequence is generated, and the abnormal probability value is subjected to time sequence correction according to the decay correction sequence to generate a time sequence correction vector. In the window smoothing stage, the abnormal probability value is subjected to time alignment and mean filtering to ensure time sequence continuity, and a smoothed probability sequence is output. A double sliding window strategy is used to extract long-term trend change characteristics and short-term fluctuation characteristics of the smoothed probability sequence. The outer window extracts the trend of the smoothed probability sequence and aggregates the features to capture medium and long-term trend change characteristics. The inner window extracts the fluctuation of the smoothed probability sequence and suppresses noise to generate short-term fluctuation characteristics. In the time weight distribution stage, the medium and long term trend change characteristics and the short term fluctuation characteristics are performed feature fusion and weight distribution to obtain a weighted feature vector; a time decay compensation factor is used to dynamically decay and quantify the weighted feature vector to obtain a time correlation degree; when the time correlation degree is lower than a decay threshold, the weighted feature vector is determined as recent data, which needs to be given a higher time weight (such as 0.7), and when the time correlation degree is higher than the decay threshold, the weighted feature vector is determined as long-term data, which needs to be given a lower time weight (such as 0.3); according to the distributed time weight, linear combination and dynamic adjustment are performed on the weighted feature vector to obtain a time weighted sequence, and the time weighted sequence is subjected to moving average processing to output a decay corrected sequence. It should be noted that the time decay compensation factor is defined based on the dynamic decay characteristics of the abnormal probability value, and the value range is [0.1, 0.9]; the decay threshold is defined based on the historical distribution characteristics of the weighted feature vector, and the value range is [0.4, 0.6].

[0036] According to the decay corrected sequence, the time sequence correction is performed on the abnormal probability value, and further, the trend decomposition is performed on the decay corrected sequence to obtain a trend feature parameter, the trend feature parameter is dynamically weighted and normalized to generate a trend correction value; according to the trend correction value, the deviation compensation is performed on the abnormal probability value to obtain a corrected probability value; the corrected probability value is subjected to smoothing filtering and nonlinear transformation to generate a time sequence correction vector.

[0037] S2.4, the time sequence correction vector is executed region division and feature aggregation to obtain a spatial feature set, and the spatial feature set is subjected to spatial correlation mapping to generate a space-time anomaly index vector, in the specific operation, the time sequence correction vector is subjected to similarity measurement to form an initial correlation matrix, the initial correlation matrix is subjected to symmetrization processing to obtain an adjacency weight matrix; the adjacency weight matrix is subjected to feature propagation and weight aggregation to construct a spatial relationship graph; the spatial relationship graph is subjected to region division by using a spectral decomposition method, further, the spatial relationship graph is subjected to eigenvalue decomposition and projection transformation to obtain a partition boundary, the partition boundary is subjected to smoothing processing and weight adjustment to generate an optimized partition, and the optimized partition is subjected to region merging to obtain an initial spatial partition; The initial spatial partition is subjected to feature aggregation, further, the initial spatial partition is subjected to multi-scale decomposition to obtain multi-level spatial features, the multi-level spatial features are subjected to texture feature extraction and geometric feature description to generate a region feature vector, the region feature vector is subjected to variance comparison and trend matching to obtain a region feature set, and the region feature set is subjected to spatial interpolation and feature fusion to output a spatial feature set; the spatial feature set is subjected to multi-dimensional feature projection to obtain a potential spatial representation, and the potential spatial representation is subjected to spatial correlation mapping to obtain a space-time anomaly index vector.

[0038] S3, extract the state features and the correlation features of the spatio-temporal anomaly index vector, and map the state features to the nodes of the topology structure and the correlation features to the edges of the topology structure to construct a dynamic causal graph.

[0039] Specifically, the method comprises the following operations, S3.1, analyze the correlation strength of the spatio-temporal anomaly index vector to form a correlation strength matrix, and perform singular value decomposition on the correlation strength matrix to extract the state features and the correlation features. In the correlation strength analysis stage, the covariance measurement is performed on the spatio-temporal anomaly index vector by using the Pearson correlation coefficient to generate a covariance matrix, and the covariance matrix is normalized by standard deviation to obtain a linear correlation coefficient. Then, polynomial fitting is performed on the linear correlation coefficient, and further, the linear correlation coefficient is discretely sampled to obtain fitting basic parameters. The fitting basic parameters are subjected to weight distribution and B-spline interpolation to obtain a smooth intensity feature curve. The smooth intensity feature curve is normalized to generate an original correlation strength distribution. At the same time, the original correlation strength distribution is subjected to dynamic weighting and dimension fusion to output an initial correlation strength vector. The initial correlation strength vector is subjected to regularization processing and matrix conversion to eliminate noise interference and ensure stability, thereby forming the correlation strength matrix. It should be noted that the Pearson correlation coefficient is defined based on the linear correlation of the spatio-temporal anomaly index vector, and the value range is [-1, 1].

[0040] In the singular value decomposition stage, the Lanczos algorithm is used to perform eigenvalue decomposition on the correlation strength matrix. Further, the first k eigenvalues of the correlation strength matrix are extracted and subjected to dimension reduction sorting to obtain a significant feature set. The significant feature set is subjected to orthogonal decomposition to generate left singular vectors and right singular vectors. The left singular vectors are subjected to rotation optimization to enhance the orthogonality of the left singular vectors, thereby obtaining optimized state vectors. The principal component projection is performed on the optimized state vectors to obtain the state features. Then, the right singular vectors are subjected to spatial standardization processing to ensure the scale consistency of the right singular vectors, thereby outputting standardized correlation vectors. The standardized correlation vectors are subjected to weighted aggregation to generate the correlation features. The state features reflect the time-varying law of the overall operation mode of the ore pulp sampler, and the correlation features represent the coupling strength between different spatial units of the ore pulp sampler.

[0041] S3.2, perform dimension normalization on the state features to obtain standard state parameters, and perform node attribute mapping on the standard state parameters to generate the nodes of the topology structure. In the dimension normalization stage, the state features are subjected to feature alignment, and the MinMax normalization is used for dimension normalization processing. At the same time, the linear transformation is used to map the dimension feature values of the state features to the interval [0, 1], thereby eliminating the dimensional difference and retaining the original distribution characteristics, and obtaining the standard state parameters. In the node attribute mapping stage, the equipment space is positioned according to the mechanical movement cutting parameters, and further, the mechanical movement cutting parameters are dynamically weighted to obtain weighted cutting parameters; the weighted cutting parameters are subjected to coordinate conversion and scale normalization to obtain equipment space coordinates; the standard state parameters and the equipment space coordinates are subjected to dimension alignment and feature splicing to obtain a fusion feature matrix; principal component analysis is applied to perform dimension scaling and orthogonal rotation on the fusion feature matrix to obtain node attribute parameters, which are subjected to spatial distribution mapping through spatial interpolation to obtain nodes of the topological structure.

[0042] S3.3, the interaction of the associated features is quantified to generate an edge weight matrix, the edge weight matrix is subjected to edge strength mapping to obtain edges of the topological structure, in specific operations, the interaction of the associated features is quantified, and further, the associated features are subjected to distribution alignment and entropy fusion to generate an associated coupling matrix, the associated coupling matrix is subjected to feature decomposition to obtain an interaction spectrum, the interaction spectrum is subjected to feature rotation and dimension compression to generate an initial mutual information matrix, and the initial mutual information matrix is subjected to dynamic weighting and scale normalization to obtain an interaction strength distribution, and the interaction strength distribution is subjected to sparse processing to obtain the edge weight matrix; In the edge strength mapping stage, the Euclidean distance of the equipment space coordinates is measured to obtain a position difference matrix, and the position difference matrix is subjected to Gaussian kernel transformation to obtain the actual physical distance of the equipment (such as pipe length, signal transmission loss); the actual physical distance of the equipment is subjected to logarithmic compression to form a standard distance vector, the standardized distance vector and the edge weight matrix are subjected to Hadamard integration to output a weighted association matrix, and a distance attenuation factor is applied to the weighted association matrix to perform nonlinear scaling, so that the weight value and the physical space are scaled uniformly, and the edge strength parameter is output, and the edge strength parameter is subjected to spatial propagation mapping to generate edges of the topological structure; It should be noted that the distance attenuation factor is defined based on the spatial distribution characteristics of the weighted association matrix, and the value range is [0.5, 0.9].

[0043] S3.4, dynamic hierarchical clustering is performed on the nodes of the topological structure and the edges of the topological structure to construct a dynamic causal graph, in specific operations, density clustering is performed on the nodes of the topological structure to generate node clustering features; the node clustering features are subjected to weight distribution and dimension aggregation to output node clustering results; at the same time, the edges of the topological structure are subjected to weight propagation and neighborhood aggregation to generate edge association features, and the edge association features are subjected to nonlinear transformation and scale compression to obtain a standardized causal matrix; the node clustering results and the standardized causal matrix are subjected to dynamic hierarchical clustering, and further, the node clustering results and the standardized causal matrix are subjected to similarity fusion to output a hierarchical feature matrix; the hierarchical feature matrix is subjected to agglomerative clustering and pruning optimization to generate an initial causal graph, and the initial causal graph is subjected to topological simplification and noise filtering to generate a dynamic causal graph.

[0044] S4, input the dynamic causal graph into a spatio-temporal convolutional neural network model, a feature aggregation layer performs node feature extraction and spatial relationship fusion, a spatio-temporal coupling layer performs multi-scale feature enhancement, and an abnormality monitoring report is output.

[0045] Specifically, the following operations are included, S4.1, constructing and training a spatio-temporal convolutional neural network model, in the specific operation, in the PyTorch framework, the GraphConv network is called through the nn.Module parameter, and the symmetric normalization is performed for the GraphConv network embedded with the adjacency normalization operator to realize the stable propagation of the node features; the input dimension of the GraphConv network is set to 64, the output dimension is set to 128, the activation function is set to ReLU, and the BatchNorm1d is connected after the GraphConv network for feature standardization, which ensures the stable propagation of the gradient, and the Dropout layer is used for random inactivation to prevent overfitting, and the feature aggregation layer is completed; the three-dimensional convolution network is called through the nn.Conv3d function, and the residual cross-layer connection mechanism is used for feature fusion to realize the stable transmission of deep features; the convolution kernel size of the three-dimensional convolution network is set to 3*3*3, the step size is set to 1*1*1, the padding mode is set to same, and the LeakyReLU function is connected after the three-dimensional convolution network for nonlinear transformation, and the spatio-temporal coupling layer is completed; The multi-head attention mechanism is used for feature interaction of the feature aggregation layer and the spatio-temporal coupling layer to obtain cross-layer attention features, the cross-layer attention features are spliced to generate a fusion feature matrix, the fusion feature matrix is mapped in dimension by using the full connection layer to obtain projection features, the projection features are normalized by using the Softmax function to generate attention weights, and the feature aggregation layer and the spatio-temporal coupling layer are stacked in cross-layer parameterization according to the attention weights, and the features are enhanced through the residual connection to complete the construction of the spatio-temporal convolutional neural network model; Next, the spatio-temporal convolutional neural network model is trained, further, the node state and edge weight data of the historical dynamic causal graph are extracted and integrated into time series data, the time series data is divided into a sample set, a training set and a validation set; on the sample set, random disturbance is performed by using data enhancement, and scale unification is performed by using standardization to form an enhanced sample; on the training set, the enhanced sample is updated in parameters by using the Adam optimizer, the early stopping mechanism is applied synchronously for training monitoring, and a validation index is obtained; on the validation set, the validation index is slidingly averaged to obtain a smoothed validation curve, the loss function is used to perform error accumulation quantification on the smoothed validation curve to obtain a loss quantization value; when the loss quantization value exceeds the convergence threshold for 5 consecutive rounds, the training is terminated, and the trained spatio-temporal convolutional neural network model is output synchronously; It should be pointed out that the convergence threshold is defined based on the dynamic change rate of the loss quantization value, and the value range is [0.001, 0.01].

[0046] S4.2, the feature aggregation layer applies an adjacency normalization operator to perform node feature extraction and spatial relationship fusion to generate a primary node embedding representation. In specific operations, the dynamic causal graph is input into the spatio-temporal convolutional neural network model through the data loader, feature alignment and weight distribution are performed on the dynamic causal graph, and initial node features are output. The initial node features are subjected to nonlinear transformation to obtain enhanced features. The adjacency normalization operator is applied to perform dynamic adjustment and range limiting on the enhanced features to obtain scale-stable features, and matrix decomposition is performed on the scale-stable features to output a normalized adjacency matrix. The normalized adjacency matrix and the initial node features are propagated and aggregated through the attention mechanism to output aggregated features. The aggregated features are subjected to spatial relationship fusion, and a full connection layer is applied for linear transformation to obtain transformed features. The transformed features are subjected to nonlinear activation using the ReLU function, and are simultaneously subjected to standardization processing through the BatchNorm layer to eliminate feature scale differences, thereby generating the primary node embedding representation. It should be pointed out that the adjacency normalization operator is defined based on the node degree distribution of the dynamic causal graph.

[0047] S4.3, the spatio-temporal coupling layer performs multi-scale feature enhancement through a residual cross-layer connection mechanism to obtain a reinforced spatio-temporal feature matrix. In specific operations, a double-layer three-dimensional convolutional network is used to perform spatio-temporal feature extraction on the primary node embedding representation. Further, the first layer of the three-dimensional convolutional network performs initial convolution and feature transformation on the primary node embedding representation, and applies LeakyReLU for nonlinear activation to generate initial spatio-temporal features. The second layer of the three-dimensional convolutional network performs secondary convolution and feature fusion on the initial spatio-temporal features to obtain deep spatio-temporal features, and performs scale normalization on the deep spatio-temporal features to obtain standard spatio-temporal features. The residual cross-layer connection mechanism is used to perform multi-scale feature enhancement on the standard spatio-temporal features. Further, channel attention calculation is performed on the standard spatio-temporal features, and weight normalization is performed through the Sigmoid function to obtain an initial weight vector. The MLP multi-layer perceptron is applied to perform nonlinear transformation and feature compression on the initial weight vector to generate channel weight values. The specific mathematical format is as follows, ; wherein, represents the channel weight value, represents the normalization factor, represents the total number of channels of the standard spatio-temporal features, represents the channel index, represents the initial weight vector of the th channel. It should be noted that the normalization factor is defined based on the output characteristics of the Sigmoid function, and the value range is (0, 1).

[0048] According to the channel weight value, the standard space-time feature is executed feature weighting and cross-layer fusion to generate a weighted space-time feature, the weighted space-time feature is executed residual connection to obtain an enhanced space-time feature; the enhanced space-time feature is executed step convolution to realize hierarchical feature extraction, and at the same time, the receptive field is expanded through the hollow convolution to output a multi-scale space-time feature; the multi-scale space-time feature is executed standardization and feature scaling using LayerNorm to output a reinforced space-time feature matrix.

[0049] S4.4, the primary node embedding representation and the reinforced space-time feature matrix are executed full connection transformation and probability normalization to form an abnormal influence degree score, the abnormal influence degree score is graded and structured integrated to output an abnormal monitoring report, in the full connection transformation stage, the primary node embedding representation and the reinforced space-time feature matrix are executed feature splicing to generate a multi-dimensional fusion feature, the multi-dimensional fusion feature is executed full connection transformation to obtain a transformed feature, and the transformed feature is executed nonlinear activation to obtain a high-dimensional feature representation; in the probability normalization stage, the high-dimensional feature representation is executed random sampling using 1*1*1 convolution to generate a probability distribution parameter, the probability distribution parameter is executed Gaussian smoothing to obtain a normalized probability, and the normalized probability is executed weighted aggregation to generate an abnormal influence degree score; The abnormal influence degree score is graded by a multi-level threshold, further, the multi-level threshold is defined according to the statistical distribution of the abnormal influence degree score, for example, the range of the first-level threshold is set to [0.8, 1.0]; the range of the second-level threshold is set to [0.6, 0.8); the range of the third-level threshold is set to [0.4, 0.6); when the abnormal influence degree score is in the range of the first-level threshold, it is defined as an emergency exception and needs to be handled immediately; when the abnormal influence degree score is in the range of the second-level threshold, it is defined as a serious exception and needs to be handled preferentially; when the abnormal influence degree score is in the range of the third-level threshold, it is defined as a general exception and needs to be handled regularly; the graded levels are executed priority sorting to output a graded exception result; In the structured integration stage, the graded exception result is executed feature alignment and weight distribution to output an abnormal feature matrix; the abnormal feature matrix is executed dimension compression and information fusion to obtain an abnormal quantization parameter, the obtained abnormal quantization parameter is executed structured layout to output an abnormal monitoring report.

[0050] The embodiment also provides an online abnormal monitoring system for a linear moving cutting ore slurry sampler, comprising: The data analysis module is configured to execute band energy separation and sliding window statistical analysis on the ore slurry sampler working condition data set, obtain a multi-dimensional feature matrix, and execute weight distribution and dynamic weighted aggregation on the multi-dimensional feature matrix to form a space-time analysis data packet. The abnormality quantification module is configured to perform a synergy analysis on the spatio-temporal analysis data packet to output a trend synergy interaction matrix, perform risk quantification and contribution degree allocation on the trend synergy interaction matrix by using an entropy weight method to generate an abnormality quantification parameter, and perform confidence weight calculation on the abnormality quantification parameter to form an abnormality probability value; and the abnormality probability value is subjected to time sequence correction and spatial correlation mapping by using a time decay compensation factor to obtain a spatio-temporal abnormality index vector. The graph construction module is configured to perform correlation strength analysis and singular value decomposition on the spatio-temporal abnormality index vector to extract state features and correlation features, map the state features to nodes of a topology structure, map the correlation features to edges of the topology structure, and construct a dynamic causal graph. The report generation module is configured to input the dynamic causal graph into a spatio-temporal convolutional neural network model, apply an adjacency normalization operator to a feature aggregation layer to perform node feature extraction and spatial relationship fusion, perform multi-scale feature enhancement by a residual cross-layer connection mechanism on a spatio-temporal coupling layer to form an abnormality influence degree score, and perform grade division and structured integration on the abnormality influence degree score to output an abnormality monitoring report.

[0051] The embodiment also provides a computer device suitable for the linearly moving cutting ore pulp sampler online abnormality monitoring method, which comprises a memory and a processor; the memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the linearly moving cutting ore pulp sampler online abnormality monitoring method proposed in the above embodiment.

[0052] The computer device can be a terminal, which comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is configured to perform wired or wireless communication with external terminals. The wireless communication can be achieved by WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0053] The embodiment also provides a storage medium on which a computer program is stored, the program being executed by a processor to implement the method for monitoring online abnormality of the linearly moving cutting ore pulp sampler proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0054] To sum up, the present application can more fully integrate multi-source data by statistical analysis of a sliding window and processing of an ore pulp sampler working condition data set by an entropy weight method, thereby accurately capturing the interaction and potential correlation between different source data. Deep feature mining and spatial relationship fusion are performed by using a dynamic causal graph and a spatiotemporal convolutional neural network model, the complex causal relationship of the ore pulp sampler under complex working conditions is effectively described, the recognition ability for complex fault modes is significantly improved, and the reliability of the abnormality monitoring is further improved.

[0055] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application 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 application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A method for online anomaly monitoring of a linearly moving cutpoint slurry sampler, the method comprising: include, Perform frequency band energy separation and sliding window statistical analysis on the slurry sampler working condition data set to obtain a multidimensional feature matrix. Then perform weight assignment and dynamic weighted aggregation on the multidimensional feature matrix to form a spatiotemporal analysis data package. Perform synergy analysis on spatiotemporal analysis data packets and output a trend synergy interaction matrix. Use the entropy weight method to perform risk quantification and contribution allocation on the trend synergy interaction matrix to generate anomaly quantification parameters. Confidence-weighted calculations are then performed on the anomaly quantification parameters to form anomaly probability values. Use the time decay compensation factor to perform time series correction and spatial correlation mapping on the anomaly probability value to obtain the spatiotemporal anomaly indicator vector; Perform correlation strength analysis and singular value decomposition on the spatiotemporal anomaly indicator vector to extract state features and correlation features. Then, map the state features into nodes of the topological structure and the correlation features into edges of the topological structure to construct a dynamic causal graph. The dynamic causal graph is input into the spatiotemporal convolutional neural network model. The feature aggregation layer applies the adjacency normalization operator to perform node feature extraction and spatial relationship fusion. The spatiotemporal coupling layer performs multi-scale feature enhancement through the residual cross-layer connection mechanism to form an anomaly impact score, which is then graded and structured to output an anomaly monitoring report.

2. The method for online abnormality monitoring of a linear moving cutting slurry sampler according to claim 1, characterized in that: The slurry sampler working condition data set includes mechanical movement cutting parameters, slurry physical property parameters and slurry sampler electrical parameters.

3. The linearly moving cutpoint slurry sampler online anomaly monitoring method of claim 1, wherein: The forming of the spatiotemporal analysis data packet specifically includes the following steps: Fast Fourier transform is applied to perform frequency band energy separation on the slurry sampler working condition data set to form a time-frequency composite parameter sequence; sliding window statistical analysis is performed on the time-frequency composite parameter sequence to obtain a multidimensional feature matrix; Perform weight assignment and vector reconstruction on the multidimensional feature matrix to generate weighted space-time vectors; perform dynamic weighted aggregation on the weighted space-time vectors and output space-time analysis data packets.

4. The linearly moving cutpoint slurry sampler online anomaly monitoring method of claim 3, wherein: The step of obtaining the spatiotemporal anomaly indicator vector specifically includes the following steps: The abnormal probability value is smoothed by window and time weight is assigned using the time decay compensation factor to generate an attenuation correction sequence. The abnormal probability value is then time-series corrected according to the attenuation correction sequence to generate a time-series correction vector. Perform regional division and feature aggregation on the time series correction vector to obtain a spatial feature set, and perform spatial correlation mapping on the spatial feature set to generate a spatiotemporal anomaly indicator vector.

5. The linearly moving cutpoint slurry sampler online anomaly monitoring method of claim 1, wherein: The construction of the dynamic causal graph specifically includes the following steps: Perform correlation strength analysis on the spatiotemporal anomaly indicator vector to form a correlation strength matrix, and perform singular value decomposition on the correlation strength matrix to extract state features and correlation features; Normalize the state features to obtain standard state parameters, perform node attribute mapping on the standard state parameters, and generate nodes of the topological structure; Quantify the interaction of the associated features to generate an edge weight matrix, perform edge strength mapping on the edge weight matrix, and obtain the edges of the topological structure; Perform dynamic hierarchical clustering on the nodes and edges of the topological structure to construct a dynamic causal graph.

6. The linearly moving cutpoint slurry sampler online anomaly monitoring method of claim 1, wherein: The spatiotemporal convolutional neural network model is constructed by building a feature aggregation layer and a spatiotemporal coupling layer, and applying a multi-head attention mechanism for cross-layer parameterized stacking.

7. The linearly moving cutpoint slurry sampler online anomaly monitoring method of claim 6, wherein: The forming abnormal influence degree score specifically comprises the following steps, The dynamic causal graph is input into a spatio-temporal convolutional neural network model, a feature aggregation layer applies an adjacency normalization operator to perform node feature extraction and spatial relationship fusion, and a primary node embedding representation is generated; A spatio-temporal coupling layer performs multi-scale feature enhancement through a residual cross-layer connection mechanism to obtain a reinforced spatio-temporal feature matrix; The primary node embedding representation and the reinforced spatio-temporal feature matrix are subjected to full connection transformation and probability normalization to form the abnormal influence degree score.

8. An online abnormality monitoring system for a linearly moving cutting ore pulp sampler, based on the online abnormality monitoring method for a linearly moving cutting ore pulp sampler according to any one of claims 1 to 7, characterized by: It comprises, A data analysis module is configured to perform band energy separation and sliding window statistical analysis on the ore pulp sampler working condition data set, obtain a multi-dimensional feature matrix, and perform weight distribution and dynamic weighted aggregation on the multi-dimensional feature matrix to form a spatio-temporal analysis data package; An abnormal quantification module is configured to perform cooperativity analysis on the spatio-temporal analysis data package to output a trend synergistic interaction matrix; perform risk quantification and contribution degree distribution on the trend synergistic interaction matrix using an entropy weight method to generate abnormal quantification parameters, and perform confidence weighted calculation on the abnormal quantification parameters to form an abnormal probability value; A time decay compensation factor is used to perform time sequence correction and spatial correlation mapping on the abnormal probability value to obtain a spatio-temporal anomaly index vector; A graph construction module is configured to perform correlation strength analysis and singular value decomposition on the spatio-temporal anomaly index vector, extract state features and correlation features, map the state features to nodes of a topology structure, map the correlation features to edges of the topology structure, and construct a dynamic causal graph; A report generation module is configured to input the dynamic causal graph into a spatio-temporal convolutional neural network model, apply an adjacency normalization operator to a feature aggregation layer to perform node feature extraction and spatial relationship fusion, perform multi-scale feature enhancement through a residual cross-layer connection mechanism in a spatio-temporal coupling layer, form an abnormal influence degree score, and perform grade division and structured integration on the abnormal influence degree score to output an abnormal monitoring report. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the linear moving cutting ore pulp sampler online abnormal monitoring method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the linear moving cutting ore pulp sampler online abnormal monitoring method of any one of claims 1-7.

Images