Equipment predictive maintenance monitoring method and electronic equipment for mine safety

Abstract

The present invention relates to the field of artificial intelligence technology, and in particular, provides a predictive maintenance monitoring method and electronic equipment for mine safety. The method includes expanding the annotated operating data to obtain an expanded data set; training a feature extraction model through the expanded data set to obtain a trained feature extraction model and feature extracted data; inputting the feature extracted data into a feature dimension reduction model to train the feature dimension reduction model to obtain a trained feature dimension reduction model and feature reduced data; training a classifier model using the feature reduced data to obtain a trained classifier model; processing new samples through the trained feature extraction model, feature dimension reduction model and classifier model to predict the operation of mine equipment. The method effectively performs predictive maintenance on mine equipment, reduces equipment failure rate, and ensures safe operation of equipment.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a mine safety-oriented equipment predictive maintenance monitoring method and electronic equipment. Background Art

[0002] With the widespread use of mining machinery and equipment, the safety and reliability of equipment have become crucial factors in mining production. Mining equipment usually operates for a long time in a complex and harsh environment, facing a variety of unfavorable conditions such as high temperature, high pressure, and strong vibration, resulting in a high risk of equipment failure. Once the equipment fails, it will not only affect the normal production of the mine, but may also bring safety hazards and even cause accidents.

[0003] In the prior art, the disclosed patent documents CN202210372271.4, CN202011592545.8 and CN201611245269.1 record technical solutions for improving the accuracy of equipment fault diagnosis, but there are still shortcomings: the traditional SMOTE algorithm can only perform linear interpolation between samples of a few categories, and cannot fully cope with the diversity and complexity of data distribution, resulting in the lack of representativeness of the generated samples and insufficient model generalization ability; the neural network model is prone to problems such as gradient vanishing or gradient explosion during the feature extraction process, and its evolution process lacks flexibility, making it difficult to adaptively adjust the model according to data characteristics, resulting in unstable training and degraded model performance; dimensionality reduction technology often causes the loss of important features, and cannot dynamically optimize feature weights during the dimensionality reduction process, resulting in redundant feature representation after dimensionality reduction, affecting the simplicity of the model and the balance between information retention; the classifier model cannot dynamically adjust the weights of different categories when dealing with category imbalance, especially in the lack of accuracy in the identification of key fault categories, resulting in inaccurate classification results in equipment maintenance monitoring. Summary of the invention

[0004] In view of this, the present invention provides a mine safety-oriented equipment predictive maintenance monitoring method and electronic equipment, which are used to effectively perform predictive maintenance on mine equipment, reduce equipment failure rate, and ensure safe operation of equipment.

[0005] In a first aspect, the present invention provides a method for predictive maintenance monitoring of equipment for mine safety, the method comprising:

[0006] S1. Collect the operation data of mining equipment and mark the operation data; and expand the marked operation data to obtain an expanded data set;

[0007] S2. Train the feature extraction model using the expanded data set to obtain a trained feature extraction model and feature-extracted data;

[0008] S3, inputting the feature-extracted data into a feature dimension reduction model, training the feature dimension reduction model, and obtaining a trained feature dimension reduction model and feature dimension reduction data;

[0009] S4, training the classifier model using the data after feature dimension reduction to obtain a trained classifier model;

[0010] S5. Process new samples through the trained feature extraction model, feature dimension reduction model and classifier model to predict the operation of mining equipment.

[0011] Optionally, collect the operating data of the mining equipment and annotate the operating data, including:

[0012] Vibration sensors, temperature sensors and pressure sensors are used to collect equipment operation data in real time and transmit it to the central database via wireless networks.

[0013] The marked categories include:

[0014] Normal operation means that the equipment is operating within normal operating parameters;

[0015] Maintenance warning, indicating that the equipment will soon require maintenance, but no failure has occurred yet;

[0016] Failure mode, indicating that the device has failed and requires immediate repair.

[0017] Optionally, the annotated running data is expanded to obtain an expanded data set, including:

[0018] The samples are generated by the SMOTE algorithm, based on which a non-convex optimization strategy is adopted. The sample generation process includes:

[0019] S101, initializing algorithm parameters, including the number of neighbors, the number of interpolation, and non-convex optimization parameters. These parameters jointly determine the generation strategy of synthetic samples, and the formula is:

[0020] ;

[0021] In the formula, Indicates the number of neighbors selected for each minority class sample in the feature space; represents the number of synthetic samples generated from each pair of minority class samples and their neighbors; represents the adjustment coefficient in the non-convex optimization process;

[0022] S102, identify a small number of categories from the training data and use them as the target of expansion. Let the identified minority class sample set be , the formula is:

[0023] ;

[0024] In the formula, is a sample in the dataset, For sample The corresponding class label; Represents a minority class, that is, the number of samples corresponding to its class is significantly less than that of other classes;

[0025] S103. For each selected minority class sample, find its The nearest neighbor samples are calculated by calculating the distance between samples. The distance calculation can be selected according to the characteristics of the data. Suppose each minority class sample is in the feature space. The nearest neighbor samples are , the formula is:

[0026] ;

[0027] In the formula, Represents the distance measurement function between samples, which is Euclidean distance or cosine similarity; For the indivual The nearest neighbor samples of For the indivual The nearest neighbor samples of ; For the overall data set;

[0028] S104. For each minority class sample and its neighbors, the feature value of the new sample point is calculated according to the non-convex optimization algorithm. The generation of the new sample is not only based on linear interpolation, but also adjusts the weight of each feature through a dynamic optimization process to better adapt to the essential structure of the data. Specifically, for each minority class sample and its neighbors, generate synthetic samples, the formula is:

[0029] ;

[0030] In the formula, is the generated synthetic sample; is a randomly generated interpolation coefficient between 0 and 1; represents a non-convex function, used to adjust according to the dynamic distribution of neighbors The influence of makes the interpolation more consistent with the distribution characteristics of the actual data;

[0031] S105. During the sample generation process, the algorithm monitors the quality of new samples and changes in category distribution in real time, and dynamically adjusts the generation strategy according to the set optimization goal. The adjustment method is expressed as:

[0032] ;

[0033] In the formula, To adjust the learning rate, To evaluate the loss function of the quality of synthetic samples, and Represent synthetic samples and their label sets respectively, Represents the loss function relative to The gradient of

[0034] S106, merge the generated synthetic samples into the original training set to form a new data set, the formula of which is:

[0035] ;

[0036] In the formula, The expanded dataset.

[0037] Optionally, the step S104 includes:

[0038] The formula for adjusting the interpolation coefficient by a non-convex function is:

[0039] ;

[0040] In the formula, and According to the neighbor sample set The two different statistics calculated represent the central tendency and dispersion tendency of the neighbor samples in the feature space;

[0041] in, The calculation method is as follows:

[0042] ;

[0043] In the formula, is a sample of neighbors The mean vector of ; It reflects the coefficient of variation of neighbor samples, and its formulas are:

[0044] ;

[0045] ;

[0046] In the formula, represents the L2 norm.

[0047] Optionally, the step S105 includes:

[0048] The loss function for evaluating the quality of synthetic samples is used to evaluate the quality of synthetic samples and adjust the non-convex optimization parameters. Its calculation method is expressed as:

[0049] ;

[0050] In the formula, is a synthetic sample predicted by the current model The class label of is the cross entropy loss function, which is used to measure the difference between the predicted label and the true label.

[0051] Optionally, step S2 includes:

[0052] A 6-layer fully connected neural network is used for feature extraction, and a neural network based on dynamic population evolution optimization is used as the feature extraction model; a self-correction mechanism is used to automatically adjust the evolution rules according to the characteristics of the current training data, and the probability of crossover and mutation is adjusted according to the changing trend of the loss function;

[0053] The training process of the neural network based on dynamic population evolution optimization includes:

[0054] S201. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, and each individual represents a configuration of network weights. Specifically, the population size is , for The individual initialization weights and biases are expressed as:

[0055] ;

[0056] ;

[0057] In the formula, For the The weight of the neural network corresponding to each individual, For the The bias of the neural network corresponding to each individual, Indicates The weight matrix of each individual in the initial state; Indicates The bias of each individual in the initial state; represents the initialization variance; It means the mean is 0 and the variance is Normal distribution of is a normal distribution; Set to 0.01;

[0058] S202: For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the first Individuals use their weights and biases to calculate the loss on the training data set, expressed as:

[0059] ;

[0060] In the formula, For the The loss of the neural network corresponding to each individual; Indicates the number of samples input in the current batch; represents the composite loss function, Represents a neural network model function; Indicates Characteristics of the samples; Indicates The labels of the samples;

[0061] The composite loss function contains a regularization term to increase the generalization ability of the model. The calculation method is expressed as:

[0062] ;

[0063] In the formula, is the mean square error function, is the regularization term, is the regularization parameter;

[0064] The calculation method of the regularization term is expressed as:

[0065] ;

[0066] In the formula, For the first The neural network corresponding to each individual weights;

[0067] S203. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of the individual being selected is expressed as:

[0068] ;

[0069] In the formula, Indicates The probability of an individual being selected; is the parameter that controls the selection pressure; For the The loss of the neural network corresponding to each individual; Set to 2;

[0070] S204. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0071] ;

[0072] ;

[0073] In the formula, is the crossover rate, is the weight of the neural network corresponding to the first individual selected, is the bias of the neural network corresponding to the first individual selected, is the weight of the neural network corresponding to the second individual selected, is the bias of the neural network corresponding to the second individual selected, is the weight of the neural network corresponding to the individual after the crossover operation, is the bias of the neural network corresponding to the individual after the crossover operation; Set to 0.3;

[0074] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0075] ;

[0076] ;

[0077] In the formula, represents the variance of the variation, is the weight of the neural network corresponding to the individual after the mutation operation, is the bias of the neural network corresponding to the individual after the mutation operation; Set to 0.04;

[0078] S205, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed.

[0079] Optionally, step S3 includes:

[0080] An autoencoder neural network based on feature refinement is used as a dimensionality reduction model. The autoencoder neural network based on feature refinement consists of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process; the feature adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, so that important features can be strengthened and minor features are gradually weakened, so that the reduced-dimensional feature representation is concise while retaining data information;

[0081] The training process of the autoencoder neural network based on feature refinement includes:

[0082] S301, assuming that the data input to the autoencoder neural network is , the encoder uses a multi-layer nonlinear mapping structure to map high-dimensional data to the initial low-dimensional feature space, which can be expressed as:

[0083] ;

[0084] In the formula, represents the initial low-dimensional features, is the weight matrix of the encoder, is the encoder bias vector, is the multi-layer Sigmoid activation function of the encoder;

[0085] S302, after the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weight is given to all features so that they can be gradually adjusted according to the feature contribution. The calculation method of the initial weight matrix is ​​expressed as:

[0086] ;

[0087] In the formula, is the initial weight matrix; is the feature weight vector, Each element in Initialized to the same value, indicating that all features have the same importance in the initial stage; is a function for extracting the diagonal elements of a matrix;

[0088] The adjusted features are expressed as:

[0089] ;

[0090] In the formula, is the feature representation after feature weight adjustment;

[0091] S303, the feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the previous round of features, gradually enhances the features with important influence, and gradually weakens the redundant or noise features. In the round iteration, the weight update rule is as follows:

[0092] ;

[0093] In the formula, Indicates The feature weights of the round iteration, Indicates The feature weights of the round iteration, is the learning rate of the autoencoder neural network, is the loss function of the autoencoder neural network, is the label data, Represents the gradient of the loss function with respect to the feature weight; the loss function of the autoencoder neural network is the reconstruction error loss function; Set to 0.01;

[0094] S304, in order to ensure the effectiveness of the dimensionality reduction process, the decoder remaps the low-dimensional features back to the high-dimensional space to ensure that the dimensionality reduction process does not lose important information; the reconstruction process of the decoder is expressed as:

[0095] ;

[0096] In the formula, is the reconstructed high-dimensional data, is the weight matrix of the decoder, is the bias vector of the decoder, is the multi-layer Sigmoid activation function of the decoder;

[0097] S305, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed.

[0098] Optionally, step S4 includes:

[0099] The classification algorithm based on extreme learning machine is used as the classification model, and the weight of each category is dynamically adjusted by adopting an adaptive reward function;

[0100] The training process of the adaptive reward-based extreme learning machine includes:

[0101] S401, initialize the extreme learning machine model structure, and set the number of nodes in the input layer, hidden layer and output layer of the extreme learning machine to , and ;in is the number of features, is the number of hidden layer nodes, is the number of output categories;

[0102] Depends on the dimension of the feature vector after feature dimensionality reduction, Set to 100, Depends on the number of categories annotated in the data;

[0103] S402, pre-adjusting the weight and bias of the entire hidden layer of the extreme learning machine through the optimization algorithm to better adapt to the initial distribution of the training data, assuming that the weight of the hidden layer of the extreme learning machine is , the deviation of the hidden layer of the extreme learning machine is , initialized to:

[0104] ;

[0105] ;

[0106] In the formula, Indicates the generation of a A matrix whose elements are randomly generated from a uniform or normal distribution;

[0107] S403, using the activation function to calculate the output of each sample in the hidden layer, assuming that the output of the hidden layer of the extreme learning machine is , the calculation method is expressed as:

[0108] ;

[0109] In the formula, is the input data matrix of the extreme learning machine; is a vector of all 1s, used to broadcast the bias term to each data sample; is the Sigmoid activation function;

[0110] S404. Before the output layer, the output weight is dynamically adjusted according to the importance and difficulty of the classification task, and continuously optimized through feedback learning, aiming to improve the model's ability to predict key faults. Specifically, the weight of the output layer is set to , which dynamically adjusts the weight according to the category error, expressed as:

[0111] ;

[0112] In the formula, is the weight learning rate; is the weight loss function, evaluating the actual output With the predicted output The error between For the The weight of the output layer of the iteration; For the The weight of the output layer of the iteration; For the categories of moderators;

[0113] The calculation method of weight loss function is expressed as:

[0114] ;

[0115] In the formula, is the element in the true label matrix, is the probability of the corresponding category predicted by the model;

[0116] The adjustment factor is adjusted according to the number of samples in the category or the training error of the previous cycle, making the weight update more refined and more adaptable. The calculation method is expressed as:

[0117] ;

[0118] In the formula, It is The number of samples for each category, is the total sample size;

[0119] S405. The weight of the output layer of the extreme learning machine is obtained by solving an optimization problem to ensure that the mapping relationship from the hidden layer to the output layer is more accurately reflected. The objective function of minimizing the output error is:

[0120] ;

[0121] In the formula, is the weight of the output layer of the extreme learning machine, Indicates data complexity;

[0122] In order to improve the adaptability and accuracy of the model, the calculation method of data complexity is expressed as:

[0123] ;

[0124] In the formula, It is The standard deviation of the feature, is the first Features

[0125] S406: Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed.

[0126] In a second aspect, an embodiment of the present invention provides an electronic device, comprising: one or more processors; a memory; and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions, which, when executed by the device, enable the device to execute the equipment predictive maintenance monitoring method for mine safety in the first aspect or any possible implementation of the first aspect.

[0127] In the technical solution provided by the present invention, the method includes collecting the operation data of the mining equipment and marking the operation data; and expanding the marked operation data to obtain an expanded data set; training the feature extraction model through the expanded data set to obtain the trained feature extraction model and the feature extracted data; inputting the feature extracted data into the feature dimension reduction model, training the feature dimension reduction model, and obtaining the trained feature dimension reduction model and the feature reduced data; using the feature reduced data to train the classifier model to obtain the trained classifier model; processing new samples through the trained feature extraction model, the feature dimension reduction model and the classifier model to predict the operation of the mining equipment. The method effectively performs predictive maintenance on the mining equipment, reduces the equipment failure rate, and ensures the safe operation of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0128] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. 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 paying creative work.

[0129] Figure 1 A flow chart of a mine safety-oriented equipment predictive maintenance monitoring method provided by an embodiment of the present invention;

[0130] Figure 2 A training flow chart of a neural network based on dynamic population evolution optimization provided by an embodiment of the present invention;

[0131] Figure 3 A training flow chart of an autoencoder neural network based on feature refinement provided in an embodiment of the present invention;

[0132] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0133] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0134] It should be clear that the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0135] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a", "said" and "the" used in the embodiments of the present invention are also intended to include plural forms, unless the context clearly indicates other meanings.

[0136] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0137] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0138] Figure 1 A flowchart of a method for predictive maintenance and monitoring of equipment for mine safety provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes:

[0139] S1. Collect the operation data of mining equipment and mark the operation data; and expand the marked operation data to obtain an expanded data set.

[0140] In the embodiment of the present invention, the operation data of the mining equipment is collected and the operation data is marked, including:

[0141] Vibration sensors, temperature sensors and pressure sensors are used to collect equipment operation data in real time and transmit it to the central database via wireless networks.

[0142] The marked categories include:

[0143] Normal operation means that the equipment is operating within normal operating parameters;

[0144] Maintenance warning, indicating that the equipment will soon require maintenance, but no failure has occurred yet;

[0145] Failure mode, indicating that the device has failed and requires immediate repair.

[0146] In the embodiment of the present invention, the annotated operating data is expanded to obtain an expanded data set, including:

[0147] The acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. The present invention generates samples through the SMOTE algorithm. The traditional SMOTE algorithm generates new sample points by interpolating between minority class samples to solve the problem of data imbalance. On this basis, the present invention adopts a non-convex optimization strategy to allow the algorithm to explore in a wider data distribution space and generate more representative and diverse synthetic samples. The sample generation process includes:

[0148] S101, initializing algorithm parameters, including the number of neighbors, the number of interpolation, and non-convex optimization parameters. These parameters jointly determine the generation strategy of synthetic samples, and the formula is:

[0149] ;

[0150] In the formula, Indicates the number of neighbors selected for each minority class sample in the feature space; represents the number of synthetic samples generated from each pair of minority class samples and their neighbors; represents the adjustment coefficient in the non-convex optimization process;

[0151] In one embodiment, The initial setting is 10. The initial setting is 8. Initial setting is 0.1.

[0152] S102, identify a small number of categories from the training data and use them as the target of expansion. Let the identified minority class sample set be , the formula is:

[0153] ;

[0154] In the formula, is a sample in the dataset, For sample The corresponding class label; Represents a minority class, that is, the number of samples corresponding to its class is significantly less than that of other classes;

[0155] S103. For each selected minority class sample, find its The nearest neighbor samples are calculated by calculating the distance between samples. The distance calculation can be selected according to the characteristics of the data. Suppose each minority class sample is in the feature space. The nearest neighbor samples are , the formula is:

[0156] ;

[0157] In the formula, Represents the distance measurement function between samples, which is Euclidean distance or cosine similarity; For the indivual The nearest neighbor samples of For the indivual The nearest neighbor samples of ; For the overall data set;

[0158] In one embodiment, the distance calculated is Euclidean distance or Manhattan distance.

[0159] S104. For each minority class sample and its neighbors, the feature value of the new sample point is calculated according to the non-convex optimization algorithm. The generation of the new sample is not only based on linear interpolation, but also adjusts the weight of each feature through a dynamic optimization process to better adapt to the essential structure of the data. Specifically, for each minority class sample and its neighbors, generate synthetic samples, the formula is:

[0160] ;

[0161] In the formula, is the generated synthetic sample; is a randomly generated interpolation coefficient between 0 and 1; represents a non-convex function, used to adjust according to the dynamic distribution of neighbors The influence of makes the interpolation more consistent with the distribution characteristics of the actual data;

[0162] S105. During the sample generation process, the algorithm monitors the quality of new samples and changes in category distribution in real time, and dynamically adjusts the generation strategy according to the set optimization goal. The adjustment method is expressed as:

[0163] ;

[0164] In the formula, To adjust the learning rate, To evaluate the loss function of the quality of synthetic samples, and Represent synthetic samples and their label sets respectively, Represents the loss function relative to The gradient of

[0165] According to the set optimization objectives, such as minimizing the intra-class distance and maximizing the inter-class distance, it makes the data augmentation process more intelligent and adaptive.

[0166] S106, merge the generated synthetic samples into the original training set to form a new data set, the formula of which is:

[0167] ;

[0168] In the formula, The expanded dataset.

[0169] In the embodiment of the present invention, step S104 includes:

[0170] The formula for adjusting the interpolation coefficient by a non-convex function is:

[0171] ;

[0172] In the formula, and According to the neighbor sample set The two different statistics calculated represent the central tendency and dispersion tendency of the neighbor samples in the feature space;

[0173] in, The calculation method is as follows:

[0174] ;

[0175] In the formula, is a sample of neighbors The mean vector of ; It reflects the coefficient of variation of neighbor samples, and its formulas are:

[0176] ;

[0177] ;

[0178] In the formula, represents the L2 norm.

[0179] In the embodiment of the present invention, step S105 includes:

[0180] The loss function for evaluating the quality of synthetic samples is used to evaluate the quality of synthetic samples and adjust the non-convex optimization parameters. Its calculation method is expressed as:

[0181] ;

[0182] In the formula, is a synthetic sample predicted by the current model The class label of is the cross entropy loss function, which is used to measure the difference between the predicted label and the true label.

[0183] Using traditional data augmentation techniques, such as rotation, flipping, scaling, noise injection, etc., these methods are not as effective as the method of the present invention on device data, but are simpler and easier to implement.

[0184] S2. Train the feature extraction model using the expanded data set to obtain a trained feature extraction model and feature-extracted data.

[0185] In the embodiment of the present invention, step S2 includes:

[0186] In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, they may encounter problems such as gradient vanishing, gradient exploding, or falling into local optimal solutions, which will affect the stability of training and the performance of the model. In traditional population evolution algorithms, the evolution of all individuals is based on fixed rules. The present invention uses a 6-layer fully connected neural network for feature extraction, and uses a neural network based on dynamic population evolution optimization as a feature extraction model. It uses a self-correction mechanism to automatically adjust the evolution rules according to the characteristics of the current training data, and adjusts the probability of crossover and mutation according to the changing trend of the loss function, so that the algorithm can adapt to different data distributions more flexibly, and improve the generalization ability and training efficiency of the model.

[0187] In the embodiment of the present invention, Figure 2 As shown in the figure, the training process of the neural network based on dynamic population evolution optimization includes:

[0188] S201. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, and each individual represents a configuration of network weights. Specifically, the population size is , for The individual initialization weights and biases are expressed as:

[0189] ;

[0190] ;

[0191] In the formula, For the The weight of the neural network corresponding to each individual, For the The bias of the neural network corresponding to each individual, Indicates The weight matrix of each individual in the initial state; Indicates The bias of each individual in the initial state; represents the initialization variance; It means the mean is 0 and the variance is Normal distribution of is a normal distribution; Set to 0.01;

[0192] S202: For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the first Individuals use their weights and biases to calculate the loss on the training data set, expressed as:

[0193] ;

[0194] In the formula, For the The loss of the neural network corresponding to each individual; Indicates the number of samples input in the current batch; represents the composite loss function, Represents a neural network model function; Indicates Characteristics of the samples; Indicates The labels of the samples;

[0195] The composite loss function contains a regularization term to increase the generalization ability of the model. The calculation method is expressed as:

[0196] ;

[0197] In the formula, is the mean square error function, is the regularization term, is the regularization parameter;

[0198] The calculation method of the regularization term is expressed as:

[0199] ;

[0200] In the formula, For the first The neural network corresponding to each individual weights;

[0201] S203. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of the individual being selected is expressed as:

[0202] ;

[0203] In the formula, Indicates The probability of an individual being selected; is the parameter that controls the selection pressure; For the The loss of the neural network corresponding to each individual; Set to 2;

[0204] S204. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as:

[0205] ;

[0206] ;

[0207] In the formula, is the crossover rate, is the weight of the neural network corresponding to the first individual selected, is the bias of the neural network corresponding to the first individual selected, is the weight of the neural network corresponding to the second individual selected, is the bias of the neural network corresponding to the second individual selected, is the weight of the neural network corresponding to the individual after the crossover operation, is the bias of the neural network corresponding to the individual after the crossover operation; Set to 0.3;

[0208] The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as:

[0209] ;

[0210] ;

[0211] In the formula, represents the variance of the variation, is the weight of the neural network corresponding to the individual after the mutation operation, is the bias of the neural network corresponding to the individual after the mutation operation; Set to 0.04;

[0212] S205, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed.

[0213] In one embodiment, the preset condition for stopping iteration is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0214] S3. Input the data after feature extraction into the feature dimension reduction model, train the feature dimension reduction model, and obtain the trained feature dimension reduction model and the data after feature dimension reduction.

[0215] In the embodiment of the present invention, an autoencoder neural network based on feature refinement is used as a dimensionality reduction model. The autoencoder neural network based on feature refinement consists of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process; the feature adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, so that important features can be strengthened and secondary features are gradually weakened, so that the reduced-dimensional feature representation has simplicity while retaining data information;

[0216] In the embodiment of the present invention, Figure 3 As shown in Figure 1, the training process of the autoencoder neural network based on feature refinement includes:

[0217] S301, assuming that the data input to the autoencoder neural network is , the encoder uses a multi-layer nonlinear mapping structure to map high-dimensional data to the initial low-dimensional feature space, which can be expressed as:

[0218] ;

[0219] In the formula, represents the initial low-dimensional features, is the weight matrix of the encoder, is the encoder bias vector, is the multi-layer Sigmoid activation function of the encoder;

[0220] S302, after the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weight is given to all features so that they can be gradually adjusted according to the feature contribution. The calculation method of the initial weight matrix is ​​expressed as:

[0221] ;

[0222] In the formula, is the initial weight matrix; is the feature weight vector, Each element in Initialized to the same value, indicating that all features have the same importance in the initial stage; is a function for extracting the diagonal elements of a matrix;

[0223] The adjusted features are expressed as:

[0224] ;

[0225] In the formula, is the feature representation after feature weight adjustment;

[0226] S303, the feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the previous round of features, gradually enhances the features with important influence, and gradually weakens the redundant or noise features. In the round iteration, the weight update rule is as follows:

[0227] ;

[0228] In the formula, Indicates The feature weights of the round iteration, Indicates The feature weights of the round iteration, is the learning rate of the autoencoder neural network, is the loss function of the autoencoder neural network, is the label data, Represents the gradient of the loss function with respect to the feature weight; the loss function of the autoencoder neural network is the reconstruction error loss function; Set to 0.01;

[0229] S304, in order to ensure the effectiveness of the dimensionality reduction process, the decoder remaps the low-dimensional features back to the high-dimensional space to ensure that the dimensionality reduction process does not lose important information; the reconstruction process of the decoder is expressed as:

[0230] ;

[0231] In the formula, is the reconstructed high-dimensional data, is the weight matrix of the decoder, is the bias vector of the decoder, is the multi-layer Sigmoid activation function of the decoder;

[0232] S305, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed.

[0233] In one embodiment, the preset condition for stopping iteration is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0234] S4. Train the classifier model using the data after feature dimension reduction to obtain a trained classifier model.

[0235] In the embodiment of the present invention, step S4 includes:

[0236] The classification algorithm based on extreme learning machine is used as the classification model. Extreme learning machine is a single-layer feedforward neural network, which is widely used in classification tasks due to its fast training speed and good generalization ability. In order to further improve the prediction accuracy and adaptability of the model, an adaptive reward function is used to dynamically adjust the weight of each category, especially in the predictive maintenance monitoring of mining equipment, where different categories have different importance and require high recognition accuracy of fault categories, thereby improving the accuracy of classification.

[0237] The training process of the adaptive reward-based extreme learning machine includes:

[0238] S401, initialize the extreme learning machine model structure, and set the number of nodes in the input layer, hidden layer and output layer of the extreme learning machine to , and ;in is the number of features, is the number of hidden layer nodes, is the number of output categories;

[0239] Depends on the dimension of the feature vector after feature dimensionality reduction, Set to 100, Depends on the number of categories annotated in the data;

[0240] S402. In a traditional extreme learning machine, the weights and biases of the hidden layer are usually randomly generated. The weights and biases of the entire hidden layer of the extreme learning machine are pre-adjusted by an optimization algorithm to better adapt to the initial distribution of training data. The weights of the hidden layer of the extreme learning machine are assumed to be , the deviation of the hidden layer of the extreme learning machine is , initialized to:

[0241] ;

[0242] ;

[0243] In the formula, Indicates the generation of a A matrix whose elements are randomly generated from a uniform or normal distribution;

[0244] S403, using the activation function to calculate the output of each sample in the hidden layer, assuming that the output of the hidden layer of the extreme learning machine is , the calculation method is expressed as:

[0245] ;

[0246] In the formula, is the input data matrix of the extreme learning machine; is a vector of all 1s, used to broadcast the bias term to each data sample; is the Sigmoid activation function;

[0247] S404. Before the output layer, the output weight is dynamically adjusted according to the importance and difficulty of the classification task, and continuously optimized through feedback learning, aiming to improve the model's ability to predict key faults. Specifically, the weight of the output layer is set to , which dynamically adjusts the weight according to the category error, expressed as:

[0248] ;

[0249] In the formula, is the weight learning rate; is the weight loss function, evaluating the actual output With the predicted output The error between For the The weight of the output layer of the iteration; For the The weight of the output layer of the iteration; For the categories of moderators;

[0250] The calculation method of weight loss function is expressed as:

[0251] ;

[0252] In the formula, is the element in the true label matrix, is the probability of the corresponding category predicted by the model;

[0253] The weighted loss function helps to gain a deeper understanding of how the reward function dynamically adjusts the weights based on the accuracy of the class prediction.

[0254] The adjustment factor is adjusted according to the number of samples in the category or the training error of the previous cycle, making the weight update more refined and more adaptable. The calculation method is expressed as:

[0255] ;

[0256] In the formula, It is The number of samples for each category, is the total sample size;

[0257] S405. The weight of the output layer of the extreme learning machine is obtained by solving an optimization problem to ensure that the mapping relationship from the hidden layer to the output layer is more accurately reflected. The objective function of minimizing the output error is:

[0258] ;

[0259] In the formula, is the weight of the output layer of the extreme learning machine, Indicates data complexity;

[0260] In order to improve the adaptability and accuracy of the model, the calculation method of data complexity is expressed as:

[0261] ;

[0262] In the formula, It is The standard deviation of the feature, is the first Features

[0263] S406: Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed.

[0264] In one embodiment, the preset condition for stopping iteration is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0265] S5. Process new samples through the trained feature extraction model, feature dimension reduction model and classifier model to predict the operation of mining equipment.

[0266] In an embodiment of the present invention, a trained model is used to process new samples to achieve predictive maintenance monitoring of equipment for mine safety. In one embodiment, the collected raw data is input into a trained feature extraction and feature dimension reduction model for feature processing, and further, the processed features are input into a classifier model for classification to obtain classification results. In an embodiment of the present invention, the classification categories include: normal operation, indicating that the equipment is operating within the normal operating parameter range; maintenance warning, indicating that the equipment is about to require maintenance, but no failure has occurred; failure mode, indicating that the equipment has failed and needs immediate repair.

[0267] Compared with the prior art, the present invention has the following beneficial effects:

[0268] 1. The SMOTE algorithm based on non-convex optimization expands the distribution range of data and solves the problem of unbalanced data samples. The traditional SMOTE algorithm only interpolates between a few classes of samples to generate new samples. The present invention uses a non-convex optimization strategy to explore in a wider data distribution space and generate more representative and diverse samples, thereby improving the generalization ability of the model and improving the model training bias problem caused by data imbalance.

[0269] 2. The neural network based on dynamic population evolution optimization makes the feature extraction process more flexible, solving problems such as gradient vanishing and gradient explosion that may occur in traditional neural network training. By dynamically adjusting the evolution rules and adaptively adjusting the probabilities of crossover and mutation operations according to data characteristics, the training stability and efficiency of the feature extraction model are improved.

[0270] 3. The feature dimensionality reduction is performed using an autoencoder neural network based on feature refinement, which solves the problem of feature information loss during the dimensionality reduction process. Through the feature adjustment module, the feature weights are dynamically adjusted and the feature representation is recursively optimized to strengthen important features while weakening minor features, thus ensuring the simplicity and integrity of the features after dimensionality reduction.

[0271] 4. Classifier training is performed through extreme learning machines based on adaptive reward functions, which improves the classifier's ability to identify key fault categories, solves the problem of uneven weight distribution of different categories in predictive maintenance of mining equipment, and dynamically adjusts the weight of the output layer. In particular, the accuracy of the model is enhanced in fault category identification, which effectively improves the accuracy of equipment status classification and improves the accurate judgment of faults and maintenance warnings.

[0272] Each step of the embodiment of the present invention may be performed by an electronic device, which includes but is not limited to a mobile phone, a tablet computer, a portable PC, a desktop computer, etc.

[0273] In the technical solution provided by the present invention, the method includes collecting the operation data of the mining equipment and marking the operation data; and expanding the marked operation data to obtain an expanded data set; training the feature extraction model through the expanded data set to obtain the trained feature extraction model and the feature extracted data; inputting the feature extracted data into the feature dimension reduction model, training the feature dimension reduction model, and obtaining the trained feature dimension reduction model and the feature reduced data; using the feature reduced data to train the classifier model to obtain the trained classifier model; processing new samples through the trained feature extraction model, the feature dimension reduction model and the classifier model to predict the operation of the mining equipment. The method effectively performs predictive maintenance on the mining equipment, reduces the equipment failure rate, and ensures the safe operation of the equipment.

[0274] An embodiment of the present invention provides a computer-readable storage medium, which includes a stored program, wherein when the program is running, the electronic device where the computer-readable storage medium is located is controlled to execute the above-mentioned embodiment of the equipment predictive maintenance monitoring method for mine safety.

[0275] Figure 4 A schematic diagram of an electronic device provided by an embodiment of the present invention, such as Figure 4 As shown, the electronic device 21 includes: a processor 211, a memory 212, and a computer program 213 stored in the memory 212 and executable on the processor 211. When the computer program 213 is executed by the processor 211, the equipment predictive maintenance monitoring method for mine safety in the embodiment is implemented. To avoid repetition, they are not described one by one here.

[0276] The electronic device 21 includes, but is not limited to, a processor 211 and a memory 212. Those skilled in the art will appreciate that Figure 4 It is only an example of the electronic device 21 and does not constitute a limitation of the electronic device 21. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0277] The processor 211 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0278] The memory 212 may be an internal storage unit of the electronic device 21, such as a hard disk or memory of the electronic device 21. The memory 212 may also be an external storage device of the electronic device 21, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card (FlashCard), etc. equipped on the electronic device 21. Further, the memory 212 may also include both an internal storage unit of the electronic device 21 and an external storage device. The memory 212 is used to store computer programs and other programs and data required by network devices. The memory 212 may also be used to temporarily store data that has been output or is to be output.

[0279] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0280] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A predictive maintenance monitoring method for equipment for mine safety, characterized in that: The method comprises: S1. Collect the operation data of mining equipment and mark the operation data; and expand the marked operation data to obtain an expanded data set; S2. Train the feature extraction model using the expanded data set to obtain a trained feature extraction model and feature-extracted data; S3, inputting the feature-extracted data into a feature dimension reduction model, training the feature dimension reduction model, and obtaining a trained feature dimension reduction model and feature dimension reduction data; S4, training the classifier model using the data after feature dimension reduction to obtain a trained classifier model; S5, processing new samples through the trained feature extraction model, feature dimension reduction model and classifier model to predict the operation of mining equipment; Collect the operating data of mining equipment and mark the operating data, including: Vibration sensors, temperature sensors and pressure sensors are used to collect equipment operation data in real time and transmit it to the central database via wireless networks. The marked categories include: Normal operation means that the equipment is operating within normal operating parameters; Maintenance warning, indicating that the equipment will soon require maintenance, but no failure has occurred yet; Failure mode, which means that the equipment is malfunctioning and requires immediate repair; The annotated running data is expanded to obtain an expanded data set, including: The samples are generated by the SMOTE algorithm, based on which a non-convex optimization strategy is adopted. The sample generation process includes: S101, initializing algorithm parameters, including the number of neighbors, the number of interpolation, and non-convex optimization parameters. These parameters jointly determine the generation strategy of synthetic samples, and the formula is: ； In the formula, Indicates the number of neighbors selected for each minority class sample in the feature space; represents the number of synthetic samples generated from each pair of minority class samples and their neighbors; represents the adjustment coefficient in the non-convex optimization process; S102, identify a small number of categories from the training data and use them as the target of expansion. Let the identified minority class sample set be , the formula is: ； In the formula, is a sample in the dataset, For sample The corresponding class label; Represents a minority class, that is, the number of samples corresponding to its class is significantly less than that of other classes; S103. For each selected minority class sample, find its The nearest neighbor samples are calculated by calculating the distance between samples. The distance calculation can be selected according to the characteristics of the data. Suppose each minority class sample is in the feature space. The nearest neighbor samples are , the formula is: ； In the formula, Represents the distance measurement function between samples, which is Euclidean distance or cosine similarity; For the indivual The nearest neighbor samples of For the indivual The nearest neighbor samples of For the overall data set; S104. For each minority class sample and its neighbors, the feature value of the new sample point is calculated according to the non-convex optimization algorithm. The generation of the new sample is not only based on linear interpolation, but also adjusts the weight of each feature through a dynamic optimization process to better adapt to the essential structure of the data. Specifically, for each minority class sample and its neighbors, generate synthetic samples, the formula is: ； In the formula, is the generated synthetic sample; is a randomly generated interpolation coefficient between 0 and 1; represents a non-convex function, used to adjust according to the dynamic distribution of neighbors The influence of makes the interpolation more consistent with the distribution characteristics of the actual data; S105. During the sample generation process, the algorithm monitors the quality of new samples and changes in category distribution in real time, and dynamically adjusts the generation strategy according to the set optimization goal. The adjustment method is expressed as: ； In the formula, To adjust the learning rate, To evaluate the loss function of the quality of synthetic samples, and Represent synthetic samples and their label sets respectively, Represents the loss function relative to The gradient of S106, merge the generated synthetic samples into the original training set to form a new data set, the formula of which is: ； In the formula, is the expanded data set; The step S104 includes: The formula for adjusting the interpolation coefficient by a non-convex function is: ； In the formula, and According to the neighbor sample set The two different statistics calculated represent the central tendency and dispersion tendency of the neighbor samples in the feature space; in, The calculation method is as follows: ； In the formula, is a sample of neighbors The mean vector of ; It reflects the coefficient of variation of neighbor samples, and its formulas are: ； ； In the formula, represents the L2 norm.

2. The method according to claim 1, characterized in that The step S105 comprises: The loss function for evaluating the quality of synthetic samples is used to evaluate the quality of synthetic samples and adjust the non-convex optimization parameters. Its calculation method is expressed as: ； In the formula, is a synthetic sample predicted by the current model The class label of is the cross entropy loss function, which is used to measure the difference between the predicted label and the true label.

3. The method according to claim 1, characterized in that The step S2 comprises: A 6-layer fully connected neural network is used for feature extraction, and a neural network based on dynamic population evolution optimization is used as the feature extraction model; a self-correction mechanism is used to automatically adjust the evolution rules according to the characteristics of the current training data, and the probability of crossover and mutation is adjusted according to the changing trend of the loss function; The training process of the neural network based on dynamic population evolution optimization includes: S201. According to the bionic algorithm initialization method, in the initialization stage, an initial population is generated, and each individual represents a configuration of network weights. Specifically, the population size is , and Respectively The weights and biases of the neural network initialization corresponding to each individual are expressed as: ； ； In the formula, For the The weight of the neural network corresponding to each individual, For the The bias of the neural network corresponding to each individual; represents the initialization variance; It means the mean is 0 and the variance is Normal distribution of is a normal distribution; Set to 0.01; S202: For each individual in the population, use its corresponding neural network configuration to process the input training data, calculate the output of the model, and evaluate its performance according to a predetermined loss function. Specifically, for the first Individuals use their weights and biases to calculate the loss on the training data set, expressed as: ； In the formula, For the The loss of the neural network corresponding to each individual; Indicates the number of samples input in the current batch; represents the composite loss function, Represents a neural network model function; Indicates Characteristics of the samples; Indicates The labels of the samples; The composite loss function contains a regularization term to increase the generalization ability of the model. The calculation method is expressed as: ； In the formula, is the mean square error function, is the regularization term, is the regularization parameter; The calculation method of the regularization term is expressed as: ； In the formula, For the The neural network corresponding to each individual weights; S203. According to the fitness of the individual, the best performing individual is selected from the current population and retained as the candidate solution for the next generation. Specifically, based on the fitness of the individual, the excellent individual has a higher probability of being selected. The calculation method of the probability of the individual being selected is expressed as: ； In the formula, Indicates The probability of an individual being selected; is the parameter that controls the selection pressure; For the The loss of the neural network corresponding to each individual; Set to 2; S204. Generate new individuals through crossover and mutation operations. The crossover operation allows two excellent individuals to exchange some genes to generate new offspring; the mutation operation randomly changes some genes in an individual to increase the diversity of the population. Specifically, the crossover operation randomly selects two individuals for gene exchange, which is expressed as: ； ； In the formula, is the crossover rate, is the weight of the neural network corresponding to the first individual selected, is the bias of the neural network corresponding to the first individual selected, is the weight of the neural network corresponding to the second individual selected, is the bias of the neural network corresponding to the second individual selected, is the weight of the neural network corresponding to the individual after the crossover operation, is the bias of the neural network corresponding to the individual after the crossover operation; Set to 0.3; The mutation operation performs a small random perturbation on the newly generated individual weights, expressed as: ； ； In the formula, represents the variance of the variation, is the weight of the neural network corresponding to the individual after the mutation operation, is the bias of the neural network corresponding to the individual after the mutation operation; Set to 0.04; S205, repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed.

4. The method according to claim 1, characterized in that: The step S3 comprises: An autoencoder neural network based on feature refinement is used as a dimensionality reduction model. The autoencoder neural network based on feature refinement consists of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process; the feature adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, so that important features can be strengthened and minor features are gradually weakened, so that the reduced-dimensional feature representation is concise while retaining data information; The training process of the autoencoder neural network based on feature refinement includes: S301, assuming that the data input to the autoencoder neural network is , the encoder uses a multi-layer nonlinear mapping structure to map high-dimensional data to the initial low-dimensional feature space, which can be expressed as: ； In the formula, represents the initial low-dimensional features, is the weight matrix of the encoder, is the encoder bias vector, is the multi-layer Sigmoid activation function of the encoder; S302, after the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weight is given to all features so that they can be gradually adjusted according to the feature contribution. The calculation method of the initial weight matrix is ​​expressed as: ； In the formula, is the initial weight matrix; is the feature weight vector, Each element in Initialized to the same value, indicating that all features have the same importance in the initial stage; is a function for extracting the diagonal elements of a matrix; The adjusted features are expressed as: ； In the formula, is the feature representation after feature weight adjustment; S303, the feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the previous round of features, gradually enhances the features with important influence, and gradually weakens the redundant or noise features. In the round iteration, the weight update rule is as follows: ； In the formula, Indicates The feature weights of the round iteration, Indicates The feature weights of the round iteration, is the learning rate of the autoencoder neural network, is the loss function of the autoencoder neural network, is the label data, Represents the gradient of the loss function with respect to the feature weight; the loss function of the autoencoder neural network is the reconstruction error loss function; Set to 0.01; S304, in order to ensure the effectiveness of the dimensionality reduction process, the decoder remaps the low-dimensional features back to the high-dimensional space to ensure that the dimensionality reduction process does not lose important information; the reconstruction process of the decoder is expressed as: ； In the formula, is the reconstructed high-dimensional data, is the weight matrix of the decoder, is the bias vector of the decoder, is the multi-layer Sigmoid activation function of the decoder; S305, repeat the above steps until the preset stop iteration condition is met, which means the model training is completed.

5. The method according to claim 1, characterized in that The step S4 comprises: The classification algorithm based on extreme learning machine is used as the classification model, and the weight of each category is dynamically adjusted by adopting an adaptive reward function; The training process of the adaptive reward-based extreme learning machine includes: S401, initialize the extreme learning machine model structure, and set the number of nodes in the input layer, hidden layer and output layer of the extreme learning machine to , and ;in is the number of features, is the number of hidden layer nodes, is the number of output categories; Depends on the dimension of the feature vector after feature dimensionality reduction, Set to 100, Depends on the number of categories annotated in the data; S402, pre-adjusting the weight and bias of the entire hidden layer of the extreme learning machine through the optimization algorithm to better adapt to the initial distribution of the training data, assuming that the weight of the hidden layer of the extreme learning machine is , the deviation of the hidden layer of the extreme learning machine is , initialized to: ； ； In the formula, Indicates the generation of a A matrix whose elements are randomly generated from a uniform or normal distribution; S403, using the activation function to calculate the output of each sample in the hidden layer, assuming that the output of the hidden layer of the extreme learning machine is , the calculation method is expressed as: ； In the formula, is the input data matrix of the extreme learning machine; is a vector of all 1s, used to broadcast the bias term to each data sample; is the Sigmoid activation function; S404. Before the output layer, the output weight is dynamically adjusted according to the importance and difficulty of the classification task, and continuously optimized through feedback learning, aiming to improve the model's ability to predict key faults. Specifically, the weight of the output layer is set to , which dynamically adjusts the weight according to the category error, expressed as: ； In the formula, is the weight learning rate; is the weight loss function, evaluating the actual output With the predicted output The error between For the The weight of the output layer of the iteration; For the The weight of the output layer of the iteration; For the categories of moderators; The calculation method of weight loss function is expressed as: ； In the formula, is the element in the true label matrix, is the probability of the corresponding category predicted by the model; The adjustment factor is adjusted according to the number of samples in the category or the training error of the previous cycle, making the weight update more refined and more adaptable. The calculation method is expressed as: ； In the formula, It is The number of samples for each category, is the total sample size; S405. The weight of the output layer of the extreme learning machine is obtained by solving an optimization problem to ensure that the mapping relationship from the hidden layer to the output layer is more accurately reflected. The objective function of minimizing the output error is: ； In the formula, is the weight of the output layer of the extreme learning machine, Indicates data complexity; In order to improve the adaptability and accuracy of the model, the calculation method of data complexity is expressed as: ； In the formula, It is The standard deviation of the feature, is the first Features S406: Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed.

6. An electronic device, characterized in that: include: one or more processors; Memory; and one or more computer programs, wherein the one or more computer programs are stored in the memory, and the one or more computer programs include instructions, which, when executed by the device, enable the device to execute the equipment predictive maintenance monitoring method for mine safety as described in any one of claims 1 to 5.

Images