A method for obtaining a maintenance strategy of a fully mechanized support

Abstract

This invention discloses a method for obtaining maintenance strategies for fully mechanized mining supports in the field of maintenance technology. It aims to solve the technical problem that the lack of scientific quantification in the acquisition process of maintenance strategies leads to the inability to promptly repair hidden dangers. The method includes: using a sequence of working parameters with actual physical meaning as the primary information source to obtain dimensional and temporal features; using these two types of features to search for fault traces from two completely different perspectives, resulting in a more accurate fault type and its probability of occurrence; additionally introducing the age of the fully mechanized mining support, which is equally important to the occurrence of faults and difficult to reflect from physical information, to obtain accurate maintenance costs and steps for a specific age; based on the above information, maintenance costs and probabilities can be used as importance references to arrange the maintenance steps corresponding to various faults, resulting in a scientifically quantified maintenance strategy that quantifies time, physical information, and cost.

Description

Technical Field

[0001] This invention relates to a method for obtaining maintenance strategies for fully mechanized mining supports, belonging to the field of fully mechanized mining support maintenance technology. Background Technology

[0002] Fully mechanized mining supports are core mechanical equipment in mining production, and their working condition is directly related to the life safety of underground workers.

[0003] Fully mechanized mining supports are characterized by high equipment value and harsh operating conditions. Without effective maintenance strategies, long-term operation of fully mechanized mining supports is prone to various faults such as hydraulic leakage, valve jamming, seal failure, and structural damage, seriously affecting normal equipment operation and posing safety hazards. Diagnosing and maintaining the working condition of fully mechanized mining supports is a critical link in ensuring safe mining operations. At present, the industry mainly adopts the "repair after an accident" and "periodic maintenance" models for fully mechanized mining supports, and maintenance decisions rely on subjective judgment based on human experience. The problems with periodic maintenance are: if the maintenance cycle is too long, the fully mechanized mining supports will lack maintenance; if the maintenance cycle is too short, maintenance costs will increase, and excessive maintenance work itself will reduce the structural life of the fully mechanized mining supports. The "repair after an accident" model is prone to long-term work stoppages, and long downtime can easily lead to the expansion of faults and secondary damage to the equipment.

[0004] Therefore, the existing process of acquiring maintenance strategies lacks scientific quantification, resulting in the inability to repair problems in a timely manner when they arise. Summary of the Invention

[0005] The purpose of this application is to overcome the shortcomings of the prior art and provide a method for obtaining maintenance strategies for fully mechanized mining supports that scientifically quantifies and matches the timing of maintenance intervention with the time of occurrence of potential hazards, as well as the computer equipment, readable storage medium and program product thereof.

[0006] To achieve the above objectives, this application employs the following technical solution:

[0007] Firstly, this application provides a method for obtaining maintenance strategies for fully mechanized mining supports, including:

[0008] Obtain the working parameter sequence of the fully mechanized mining support, and extract dimensional features and time-series features from the working parameter sequence;

[0009] The fault type and the probability of fault occurrence are obtained based on the dimensional features and the temporal features.

[0010] The maintenance cost and maintenance steps are determined based on the type of failure and the service life of the fully mechanized mining support.

[0011] The maintenance steps are arranged according to the maintenance cost and the probability of failure to obtain a maintenance strategy.

[0012] Furthermore, the process of determining the maintenance cost and maintenance steps based on the fault type and the service life of the fully mechanized mining support includes:

[0013] The maintenance time, maintenance cost, and troubleshooting difficulty value are obtained based on the fault type, the service life of the fully mechanized mining support, and the trained cost factor prediction network.

[0014] The repair cost is determined by combining the repair time, repair cost, and troubleshooting difficulty.

[0015] The maintenance steps are derived based on the service life of the fully mechanized mining support, the fault type, and a pre-constructed knowledge graph.

[0016] Furthermore, the step of obtaining the maintenance time, maintenance cost, and troubleshooting difficulty value based on the fault type, the service life of the fully mechanized mining support, and the trained cost factor prediction network includes:

[0017] The trained cost factor prediction network includes an embedding layer, a feature splicing layer, and a fully connected regression layer;

[0018] The fault type is mapped to a low-dimensional dense vector through the embedding layer, and the low-dimensional dense vector is obtained by the following formula:

[0019] ,

[0020] In the formula, It is a low-dimensional dense vector. This is the embedding layer weight matrix. For discrete category features, One-hot encoding for fault types, This is the embedding layer bias vector;

[0021] The continuous values ​​of the service life of the fully mechanized mining support are extracted, and then concatenated with the low-dimensional dense vector through the feature splicing layer to obtain a fused feature vector; the fused feature vector is obtained by the following formula:

[0022] ,

[0023] In the formula, To fuse feature vectors, This represents a continuous numerical value of seniority. This is an operation that concatenates continuous numerical values ​​of years of service with a low-dimensional dense vector.

[0024] The fully connected regression layer is used to perform a nonlinear mapping on the fused feature vector to obtain the repair time, the repair cost, and the troubleshooting difficulty value.

[0025] ,

[0026] In the formula, This is the output vector of the fully connected regression layer. For repair time, For repair costs, To check the difficulty value, This is the output layer weight matrix. This represents the number of fully connected layers in the fully connected regression layer. The first fully connected regression layer The output of a fully connected layer This is the output layer bias vector. For the counting sequence number, The first fully connected regression layer The weight matrix of each fully connected layer The first fully connected regression layer The bias vector of each fully connected layer The first fully connected regression layer The output of a fully connected layer The first fully connected regression layer The output of a fully connected layer The activation function is the activation operation. This represents the input to the fully connected regression layer.

[0027] Furthermore, the repair cost derived by integrating the repair time, repair cost, and troubleshooting difficulty includes:

[0028] The repair time, the repair cost, and the troubleshooting difficulty value are input into the trained integrated cost fusion network, which includes an input layer, a hidden layer, and an output layer.

[0029] The repair time, repair cost, and troubleshooting difficulty value are input into three neurons of the input layer to obtain the cost factor input vector; the cost factor input vector is obtained by the following formula:

[0030] ,

[0031] In the formula, This is the operation of inputting repair time, repair cost, and troubleshooting difficulty values ​​into the three neurons of the input layer, respectively. The cost factor input vector;

[0032] The hidden layer is used to obtain the feature vector of the cost factor, which is obtained by the following formula:

[0033] ,

[0034] In the formula, The output of the first hidden layer of the integrated cost fusion network, The eigenvector of the cost factor, The activation function is the activation operation. This is the weight matrix of the first hidden layer of the integrated cost fusion network. This is the weight matrix of the second hidden layer of the integrated cost fusion network. The paranoia vector of the first hidden layer of the integrated cost fusion network. This is the paranoia vector of the second hidden layer of the integrated cost fusion network;

[0035] The maintenance cost is obtained using the feature vectors of the output layer and the cost factor, and the maintenance cost is obtained by the following formula:

[0036] ,

[0037] In the formula, For the cost of repairs, This is the output layer bias vector. This is the output layer weight matrix.

[0038] Furthermore, the step of obtaining the maintenance steps based on the service life of the fully mechanized mining support, the fault type, and a pre-constructed knowledge graph includes:

[0039] The service life of the fully mechanized mining support and the fault type are concatenated into an input condition vector; the input condition vector is encoded into a latent space input vector.

[0040] The input working condition vector is matched for vector similarity in the pre-constructed knowledge graph to obtain a relevant relation vector.

[0041] Aggregate the relevant relation vectors to obtain the translation vector;

[0042] Based on the translation vector and the latent space input vector, a maintenance scheme vector is obtained; the maintenance steps are obtained by decoding the maintenance scheme vector; the maintenance scheme vector is obtained by the following formula:

[0043] ,

[0044] In the formula, For the input working condition vector, This is an operation that concatenates the service life and fault type of the fully mechanized mining support into a vector. It is a translation vector. For the correlation vector, This is a vector aggregation operation. This is used to introduce the subsequent filtering conditions. It is a set of related relation vectors. For maintenance solution vectors, The input vector is the latent space vector. This is the decoded repair scheme vector. This represents the maintenance steps extracted from the decoded maintenance plan vector.

[0045] Further, the extraction of dimensional features and temporal features from the working parameter sequence includes:

[0046] The working parameter sequence is sequentially subjected to dimension alignment, missing value imputation, and feature dimension standardization to obtain the preprocessed working parameter sequence.

[0047] Obtain a trained MBN network model, which includes a hidden layer, a liquid state layer, and a readout layer; input the preprocessed working parameter sequence into the hidden layer of the MBN network model to obtain a data temporal feature matrix;

[0048] The liquid state matrix is ​​updated step-by-step according to the data time-series feature matrix to obtain liquid state vectors at different time steps; the liquid state vectors at each time step are concatenated to obtain the liquid state matrix; the liquid state matrix is ​​as follows:

[0049] ,

[0050] In the formula, For time steps, To update to the Liquid state vector at each time step. It is a time-varying activation function. The first of the time series feature matrix of the data The eigenvectors of the row, The data temporal feature matrix is ​​mapped from the hidden layer to a fixed sparse weight matrix in the liquid state layer. To update to the Liquid state vector at each time step. The fixed sparse recursive weight matrix for the liquid state layer. A fixed bias vector is applied to the liquid state layer. For the number of time steps, The liquid state matrix, To update to the The liquid state vector at each time step, with the upper right subscript... Represents the transpose of a matrix. Represent real numbers, This represents the number of liquid neurons in the liquid state layer.

[0051] The liquid state vector in the liquid state matrix is ​​reconstructed step-by-step using the readout layer to obtain the reconstructed feature vector. The reconstructed time-series data feature matrix is ​​then obtained based on the reconstructed feature vector. The formula for obtaining the reconstructed time-series data feature matrix is ​​as follows:

[0052] ,

[0053] In the formula, For the corresponding number Reconstructed feature vectors at each time step This is the output layer weight matrix. This is the output layer bias vector. To reconstruct the feature matrix of time series data;

[0054] Based on the reconstructed time-series data feature matrix, dimensional features and time-series features are obtained.

[0055] Further, obtaining dimensional features and time-series features based on the reconstructed time-series data feature matrix includes:

[0056] The feature dimension reconstruction error is obtained based on the reconstructed time-series data feature matrix and the data time-series feature matrix. The feature dimension reconstruction error is obtained by the following formula:

[0057] ,

[0058] In the formula, For the first Reconstruction error in each feature dimension For the counting sequence number, The first of the time series feature matrix of the data The time step and the first Feature values ​​of each feature dimension To reconstruct the first feature matrix of time series data The time step and the first Feature values ​​of each feature dimension;

[0059] The dimension anomaly contribution is obtained based on the feature dimension reconstruction error, and the dimension anomaly contribution is obtained by the following formula:

[0060] ,

[0061] In the formula, For the first The contribution of dimensional anomalies in each feature dimension;

[0062] Obtain the feature dimension error threshold, and count the number of feature dimension reconstruction errors that are greater than the feature dimension error threshold to obtain the number of abnormal feature dimensions.

[0063] Traverse all feature dimension reconstruction errors, find the number corresponding to the largest feature dimension reconstruction error, and obtain the largest outlier dimension number;

[0064] Sort all feature dimension reconstruction errors, extract the top three feature dimension reconstruction errors with the largest errors from the sequence, and form a three-dimensional anomaly dimension combination feature.

[0065] The feature dimension reconstruction error, the dimension anomaly contribution, the number of anomalous feature dimensions, the maximum anomalous dimension number, and the three-dimensional anomalous dimension combination feature are used as the dimension class feature;

[0066] The time step error sequence is obtained based on the reconstructed time series data feature matrix and the data time series feature matrix. The formula for obtaining the time step error sequence is as follows:

[0067] ,

[0068] In the formula, The time step error sequence, The number of feature dimensions;

[0069] Obtain the time step error threshold, and find the starting time step from the time step error sequence that satisfies the following condition: starting from the starting time step, the time step error sequence corresponding to 3 to 5 consecutive time steps first appears to be greater than the time step error threshold; the starting time step is taken as the abnormal start time.

[0070] Extract the time step corresponding to the maximum error value from the time step error sequence to obtain the error peak time;

[0071] The slope of the reconstruction error is obtained based on the anomaly start time and the error peak time. The formula for obtaining the slope of the reconstruction error is as follows:

[0072] ,

[0073] This is it. To reconstruct the slope of the error rise, This represents the error at the peak time in the time-step error sequence. This represents the error at the abnormal starting point within the peak error time. This is the time when the error peaks. This is the time when the anomaly began;

[0074] The duration of the anomaly is obtained by detecting the time difference from the start time of the anomaly to the current detection time step.

[0075] The error decay rate is obtained based on the peak error time and the time step error sequence. The formula for obtaining the error decay rate is as follows:

[0076] ,

[0077] In the formula, For the error decay rate, This represents the error at the abnormal end time or the current detection time step. This refers to the end time of the anomaly or the current detection time step.

[0078] The variance of the error sequence fluctuation is obtained based on the duration of the anomaly and the time step error sequence. The formula for obtaining the variance of the error sequence fluctuation is as follows:

[0079] ,

[0080] In the formula, The variance of the error sequence fluctuation. The duration of the anomaly. This represents the average error value within the duration of the anomaly in the time-step error sequence. This is the abnormal termination time;

[0081] The anomaly initiation time, the reconstruction error rise slope, the error peak time, the anomaly duration, the error decay rate, and the error sequence variance are used as the time-series features.

[0082] The step of obtaining the fault type and fault occurrence probability based on the dimensional class features and the temporal class features includes:

[0083] A comprehensive statistical feature vector is constructed based on the dimensional features and the temporal features.

[0084] A Bayesian classification model is used to process the comprehensive statistical feature vector to obtain the posterior probability of each fault type, and this posterior probability is used as the probability of fault occurrence. The formula for obtaining the posterior probability is as follows:

[0085] ,

[0086] In the formula, For comprehensive statistical feature vectors, For the first The posterior probability of each fault type For the first The characteristic conditional probability of each fault type For the first The prior probability of each fault type For the first Likelihood probability of each type of failure For the first The prior probability of each fault type To and Different counting numbers, To preset the total number of fault types, and They represent the first species and first Types of faults.

[0087] Secondly, this application also provides a computer device, including a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the maintenance strategy acquisition method for the fully mechanized mining support as described in any embodiment of the first aspect are performed.

[0088] Thirdly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the maintenance strategy acquisition method for the fully mechanized mining support described in any embodiment of the first aspect.

[0089] Fourthly, this application also provides a computer program product, including a computer program / instructions, characterized in that, when the computer program / instructions are executed by a processor, they implement the steps of the maintenance strategy acquisition method for the fully mechanized mining support described in any embodiment of the first aspect.

[0090] Compared with the prior art, the beneficial effects achieved by this application are as follows:

[0091] The maintenance strategy acquisition method for fully mechanized mining supports provided in this application uses a sequence of working parameters with actual physical meaning as the primary information source to obtain dimensional and temporal features. These two types of features are used to search for fault traces from two completely different perspectives, resulting in more accurate fault types and their probabilities of occurrence. Additionally, the age of the fully mechanized mining supports, which is equally important to the occurrence of faults but difficult to reflect from physical information, is introduced to obtain accurate maintenance costs and steps for specific ages. Based on the above information, maintenance costs and probabilities can be used as important references to arrange maintenance steps corresponding to various faults to obtain maintenance strategies. This achieves the acquisition of scientifically quantified maintenance strategies that encompass time, physical information, and cost. Attached Figure Description

[0092] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0093] Figure 1 This is a flowchart of the steps for obtaining the maintenance strategy of the fully mechanized mining support provided in the embodiments of this application;

[0094] Figure 2 This is a schematic block diagram of the computer device provided in the embodiments of this application;

[0095] Figure 3 This is another flowchart of the method for obtaining maintenance strategies for fully mechanized mining supports provided in the embodiments of this application. Detailed Implementation

[0096] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0097] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0098] Example 1:

[0099] Figure 1 This is a flowchart illustrating a method for obtaining maintenance strategies for a fully mechanized mining support according to Embodiment 1 of the present invention. This flowchart merely shows the logical sequence of the methods described in this embodiment. Provided there are no conflicts, different methods may be used in other possible embodiments of the present invention. Figure 1 Complete the steps shown or described in the order indicated.

[0100] The maintenance strategy acquisition method for the fully mechanized mining support provided in this embodiment can be applied to terminals, such as any smartphone, tablet, or computer device with communication capabilities. See also... Figure 1 The method implemented in this way specifically includes the following steps:

[0101] Obtain the working parameter sequence of the fully mechanized mining support, and extract dimensional features and time-series features from the working parameter sequence;

[0102] The fault type and the probability of fault occurrence are obtained based on the dimensional features and the temporal features.

[0103] The maintenance cost and maintenance steps are determined based on the type of failure and the service life of the fully mechanized mining support.

[0104] The maintenance steps are arranged according to the maintenance cost and the probability of failure to obtain a maintenance strategy.

[0105] The sequence of operating parameters includes operating parameters recorded along the timeline by the sensor or instrument's own logs, including but not limited to current, flow rate, pressure, top beam tilt angle, base tilt angle, shield beam tilt angle, front connecting rod tilt angle, and rear connecting rod tilt angle.

[0106] In one embodiment, the sequence of working parameters can be input into the MBN network model, which outputs dimensional features and temporal features. The MBN network model is a hybrid neural network architecture specifically designed for anomaly detection of time-series sensing data from fully mechanized mining supports.

[0107] Those skilled in the art can construct the MBN network model themselves. The MBN network model uses rRBM stacking for temporal feature extraction. The MBN network model also uses a liquid state machine for high-dimensional nonlinear state expansion. The MBN network model also uses a linear readout layer for reconstruction. This model is trained on normal samples and uses the reconstruction error deviation to discover and classify faults, providing fault probability input for subsequent maintenance cost ranking.

[0108] To use computers to perform scientific qualitative and quantitative analysis of the working parameter sequence, it is necessary to analyze each physical parameter within the working parameter sequence. This embodiment considers that the physical information related to faults often has the characteristics of multi-dimensional linkage and long-term distribution. Therefore, this embodiment analyzes and classifies the information contained in the working parameter sequence from the perspective of the dimensional features and the time-series features to obtain the fault type and the probability of fault occurrence.

[0109] However, the service life of the fully mechanized mining support is also an important factor. Different service lives of fully mechanized mining supports require different maintenance costs and specific maintenance methods. This factor is difficult to reflect in the working parameter sequence that records the physical information of the operation. Therefore, we need to introduce the service life of the fully mechanized mining support together with the previously inferred fault type as an information source to obtain the maintenance cost and maintenance steps.

[0110] In this embodiment, the maintenance steps can be pre-set for different service life and fault types. Therefore, when maintaining the fully mechanized mining support, the maintenance steps can be retrieved directly by referring to the pre-set mapping table according to the service life and fault type.

[0111] Once the maintenance cost and failure probability are obtained, all failure types can be ranked based on their importance, and then maintenance steps can be arranged to obtain a maintenance strategy.

[0112] In summary, the maintenance strategy acquisition method for fully mechanized mining supports provided in this embodiment uses a sequence of working parameters with actual physical meaning as the primary information source to obtain dimensional and temporal features. These two types of features are used to search for fault traces from two completely different perspectives, resulting in more accurate fault types and their probabilities. Furthermore, the method introduces the age of the fully mechanized mining support, which is equally important to the occurrence of faults and difficult to reflect from physical information, thereby obtaining accurate maintenance costs and steps for specific ages. Based on the above information, maintenance costs and probabilities can be used as importance references to arrange maintenance steps corresponding to various faults to obtain maintenance strategies. This achieves the acquisition of scientifically quantified maintenance strategies encompassing time, physical information, and cost.

[0113] Example 2:

[0114] This embodiment provides a method for obtaining maintenance strategies for fully mechanized mining supports. This embodiment is an optimization based on Embodiment 1 to improve technical effectiveness and refine the technical solution. For details not described in this embodiment, please refer to Embodiment 1.

[0115] As one embodiment, the method for obtaining maintenance costs and maintenance steps based on the fault type and the service life of the fully mechanized mining support in Embodiment 1 includes:

[0116] Based on the fault type, the service life of the fully mechanized mining support, and the trained cost factor prediction network, the maintenance time, maintenance cost, and troubleshooting difficulty values ​​are obtained; maintenance time, maintenance cost, and troubleshooting difficulty values ​​are the most important maintenance costs and expenses recognized by the industry.

[0117] To facilitate subsequent calculations, the repair time, repair cost, and troubleshooting difficulty value are combined to obtain a unified repair cost.

[0118] The maintenance steps are derived based on the service life of the fully mechanized mining support, the fault type, and a pre-constructed knowledge graph.

[0119] In this embodiment, the knowledge graph essentially structures historical maintenance experience into a triplet model containing working condition entities, relationships, and solution entities. Therefore, given the known service life and fault types of the fully mechanized mining support, the service life and fault types of the fully mechanized mining support can be mapped to specific maintenance steps.

[0120] As one embodiment, the step of obtaining the maintenance time, maintenance cost, and troubleshooting difficulty value based on the fault type, the service life of the fully mechanized mining support, and the trained cost factor prediction network includes:

[0121] The trained cost factor prediction network includes an embedding layer, a feature splicing layer, and a fully connected regression layer;

[0122] The fault type is mapped to a low-dimensional dense vector through the embedding layer, and the low-dimensional dense vector is obtained by the following formula:

[0123] ,

[0124] In the formula, It is a low-dimensional dense vector. This is the embedding layer weight matrix. For discrete category features, One-hot encoding for fault types, This is the embedding layer bias vector;

[0125] The continuous values ​​of the service life of the fully mechanized mining support are extracted, and then concatenated with the low-dimensional dense vector through the feature splicing layer to obtain a fused feature vector; the fused feature vector is obtained by the following formula:

[0126] ,

[0127] In the formula, To fuse feature vectors, This represents a continuous numerical value of seniority. This is an operation that concatenates continuous numerical values ​​of years of service with a low-dimensional dense vector.

[0128] The fully connected regression layer is used to perform a nonlinear mapping on the fused feature vector to obtain the repair time, the repair cost, and the troubleshooting difficulty value.

[0129] ,

[0130] In the formula, This is the output vector of the fully connected regression layer. For repair time, For repair costs, To check the difficulty value, This is the output layer weight matrix. This represents the number of fully connected layers in the fully connected regression layer. The first fully connected regression layer The output of a fully connected layer This is the output layer bias vector. For the counting sequence number, The first fully connected regression layer The output of a fully connected layer The first fully connected regression layer The weight matrix of each fully connected layer The first fully connected regression layer The bias vector of each fully connected layer The first fully connected regression layer The output of a fully connected layer The activation function is the activation operation. This represents the input to the fully connected regression layer.

[0131] As one embodiment, when distinguishing fault types, a Bayesian classification model can be used for classification and probability prediction. For example, one-hot encoding of fault types can be used. Including: leaks Slow movement of the stent Jack failure Safety valve malfunction Deformation of structural components aging of seals Filter blockage , =1,2,...,7.

[0132] As one embodiment, the step of arranging the maintenance steps according to the maintenance cost and the failure probability to obtain a maintenance strategy, as mentioned in Embodiment 1, further includes:

[0133] Extract the maintenance cost from the maintenance strategy, and use a multiplication formula to comprehensively evaluate the probability-based evaluation results for each fault type:

[0134] ;

[0135] In the formula, The serial number is the unique hot code for the fault type. For the first The combined probabilistic cost of each type of failure For the first The probability of occurrence of each type of failure For the first The cost of each type of failure.

[0136] The maintenance steps in the maintenance strategy are reordered according to the principle of descending the comprehensive probability cost of the fault type to obtain the optimized maintenance strategy, which is then used as the final maintenance strategy.

[0137] As one embodiment, the repair cost obtained by integrating the repair time, repair cost, and troubleshooting difficulty includes:

[0138] The repair time, repair cost, and troubleshooting difficulty value are input into a trained integrated cost fusion network, which includes an input layer, a hidden layer, and an output layer; wherein the input layer has at least three neurons, corresponding to the repair time, the repair cost, and the troubleshooting difficulty value, respectively.

[0139] The repair time, repair cost, and troubleshooting difficulty value are input into three neurons of the input layer to obtain the cost factor input vector; the cost factor input vector is obtained by the following formula:

[0140] ,

[0141] In the formula, This is the operation of inputting repair time, repair cost, and troubleshooting difficulty values ​​into the three neurons of the input layer, respectively. The cost factor input vector;

[0142] The hidden layer is used to obtain the feature vector of the cost factor. In some network models, there are two hidden layers. The feature vector of the cost factor is obtained by the following formula:

[0143] ,

[0144] In the formula, The output of the first hidden layer of the integrated cost fusion network, The eigenvector of the cost factor, The activation function is the activation operation. This is the weight matrix of the first hidden layer of the integrated cost fusion network. This is the weight matrix of the second hidden layer of the integrated cost fusion network. The paranoia vector of the first hidden layer of the integrated cost fusion network. This is the paranoia vector of the second hidden layer of the integrated cost fusion network;

[0145] The maintenance cost is obtained using the feature vectors of the output layer and the cost factor, and the maintenance cost is obtained by the following formula:

[0146] ,

[0147] In the formula, For the cost of repairs, This is the output layer bias vector. This is the output layer weight matrix.

[0148] As one embodiment, the maintenance steps, derived from the service life of the fully mechanized mining support, the fault type, and a pre-constructed knowledge graph, include:

[0149] The service life of the fully mechanized mining support and the fault type are concatenated into an input condition vector; the input condition vector is encoded into a latent space input vector.

[0150] The input working condition vector is matched for vector similarity in the pre-constructed knowledge graph to obtain a relevant relation vector.

[0151] The relevant relationship vectors are aggregated to obtain a translation vector; the translation vector contains information mapping from the service life and fault type of the fully mechanized mining support to the maintenance steps;

[0152] Based on the translation vector and the latent space input vector, a maintenance scheme vector is obtained; the maintenance steps are obtained by decoding the maintenance scheme vector; the maintenance scheme vector is obtained by the following formula:

[0153] ,

[0154] In the formula, For the input working condition vector, This is an operation that concatenates the service life and fault type of the fully mechanized mining support into a vector. It is a translation vector. For the correlation vector, This is a vector aggregation operation. This is used to introduce the subsequent filtering conditions. It is a set of related relation vectors. For maintenance solution vectors, The input vector is the latent space vector. This is the decoded repair scheme vector. This represents the maintenance steps extracted from the decoded maintenance plan vector.

[0155] As one embodiment, the extraction of dimensional features and temporal features from the working parameter sequence includes:

[0156] The working parameter sequence is sequentially subjected to dimension alignment, missing value imputation, and feature dimension standardization to obtain the preprocessed working parameter sequence.

[0157] Obtain a trained MBN network model, which includes a hidden layer, a liquid state layer, and a readout layer; input the preprocessed working parameter sequence into the hidden layer of the MBN network model to obtain a data temporal feature matrix;

[0158] The liquid state matrix is ​​updated step-by-step according to the data time-series feature matrix to obtain liquid state vectors at different time steps; the liquid state vectors at each time step are concatenated to obtain the liquid state matrix; the liquid state matrix is ​​as follows:

[0159] ,

[0160] In the formula, For time steps, To update to the Liquid state vector at each time step. It is a time-varying activation function. The first of the time series feature matrix of the data The eigenvectors of the row, The data temporal feature matrix is ​​mapped from the hidden layer to a fixed sparse weight matrix in the liquid state layer. To update to the Liquid state vector at each time step. The fixed sparse recursive weight matrix for the liquid state layer. A fixed bias vector is applied to the liquid state layer. For the number of time steps, The liquid state matrix, To update to the The liquid state vector at each time step, with the upper right subscript... Represents the transpose of a matrix. Represent real numbers, This represents the number of liquid neurons in the liquid state layer.

[0161] The liquid state vector in the liquid state matrix is ​​reconstructed step-by-step using the readout layer to obtain the reconstructed feature vector. The reconstructed time-series data feature matrix is ​​then obtained based on the reconstructed feature vector. The formula for obtaining the reconstructed time-series data feature matrix is ​​as follows:

[0162] ,

[0163] In the formula, For the corresponding number Reconstructed feature vectors at each time step This is the output layer weight matrix. This is the output layer bias vector. To reconstruct the feature matrix of time series data;

[0164] Based on the reconstructed time-series data feature matrix, dimensional features and time-series features are obtained.

[0165] As one embodiment, obtaining dimensional features and time-series features based on the reconstructed time-series data feature matrix mentioned above includes:

[0166] The feature dimension reconstruction error is obtained based on the reconstructed time-series data feature matrix and the data time-series feature matrix. The feature dimension reconstruction error is obtained by the following formula:

[0167] ,

[0168] In the formula, For the first Reconstruction error in each feature dimension For the counting sequence number, The first of the time series feature matrix of the data The time step and the first Feature values ​​of each feature dimension To reconstruct the first feature matrix of time series data The time step and the first Feature values ​​of each feature dimension;

[0169] The dimension anomaly contribution is obtained based on the feature dimension reconstruction error, and the dimension anomaly contribution is obtained by the following formula:

[0170] ,

[0171] In the formula, For the first The contribution of dimensional anomalies in each feature dimension;

[0172] Obtain the feature dimension error threshold, and count the number of feature dimension reconstruction errors that are greater than the feature dimension error threshold to obtain the number of abnormal feature dimensions.

[0173] Traverse all feature dimension reconstruction errors, find the number corresponding to the largest feature dimension reconstruction error, and obtain the largest outlier dimension number;

[0174] Sort all feature dimension reconstruction errors, extract the top three feature dimension reconstruction errors with the largest errors from the sequence, and form a three-dimensional anomaly dimension combination feature.

[0175] The feature dimension reconstruction error, the dimension anomaly contribution, the number of anomalous feature dimensions, the maximum anomalous dimension number, and the three-dimensional anomalous dimension combination feature are used as the dimension class feature;

[0176] The time step error sequence is obtained based on the reconstructed time series data feature matrix and the data time series feature matrix. The formula for obtaining the time step error sequence is as follows:

[0177] ,

[0178] In the formula, The time step error sequence, The number of feature dimensions;

[0179] Obtain the time step error threshold, and find the starting time step from the time step error sequence that satisfies the following condition: starting from the starting time step, the time step error sequence corresponding to 3 to 5 consecutive time steps first appears to be greater than the time step error threshold; the starting time step is taken as the abnormal start time.

[0180] Extract the time step corresponding to the maximum error value from the time step error sequence to obtain the error peak time;

[0181] The slope of the reconstruction error is obtained based on the anomaly start time and the error peak time. The formula for obtaining the slope of the reconstruction error is as follows:

[0182] ,

[0183] This is it. To reconstruct the slope of the error rise, This represents the error at the peak time in the time-step error sequence. This represents the error at the abnormal starting point within the peak error time. This is the time when the error peaks. This is the time when the anomaly began;

[0184] The duration of the anomaly is obtained by detecting the time difference from the start time of the anomaly to the current detection time step.

[0185] The error decay rate is obtained based on the peak error time and the time step error sequence. The formula for obtaining the error decay rate is as follows:

[0186] ,

[0187] In the formula, For the error decay rate, This represents the error at the abnormal end time or the current detection time step. This refers to the end time of the anomaly or the current detection time step.

[0188] The variance of the error sequence fluctuation is obtained based on the duration of the anomaly and the time step error sequence. The formula for obtaining the variance of the error sequence fluctuation is as follows:

[0189] ,

[0190] In the formula, The variance of the error sequence fluctuation. The duration of the anomaly. This represents the average error value within the duration of the anomaly in the time-step error sequence. This is the abnormal termination time;

[0191] The anomaly start time, the reconstruction error rise slope, the error peak time, the anomaly duration, the error decay rate, and the error sequence fluctuation variance are used as the time series features.

[0192] The maintenance strategy acquisition method for fully mechanized mining supports provided in this embodiment enables comprehensive fault assessment, multi-objective optimization decision-making, and precise formulation of adaptive maintenance plans, thereby reducing maintenance costs, minimizing downtime losses, and extending equipment service life.

[0193] As one embodiment, the method provided in this embodiment can be executed based on a network model composed of multiple modules. Each module can be trained independently. This network model includes at least a cost factor prediction module, a comprehensive cost fusion module, and a knowledge graph module for interpretation. The cost factor prediction module is used to predict the input features of the fault, namely, the length of service, the maintenance time corresponding to the fault type, the maintenance cost, and the difficulty of fault inspection. The cost factor prediction module includes an embedding layer, a feature concatenation layer, a fully connected regression layer, and an output layer. The comprehensive cost fusion module is used to fuse the predicted cost factors to obtain the comprehensive cost. The comprehensive cost fusion module includes an input layer, a hidden layer, and an output layer. The hidden layer uses the ReLU activation function. The knowledge graph module is used to input the fault input features, namely the length of service and the fault type, into the knowledge graph for reasoning, generating maintenance suggestions and steps, including the number of maintenance personnel and whether parts need to be replaced.

[0194] The cost factor prediction module's embedding layer includes an independent embedding sublayer corresponding to the fault type. This embedding sublayer converts discrete fault types into low-dimensional dense vectors. The embedding layer mapping formula is as follows:

[0195] ;

[0196] The feature concatenation layer of the cost factor prediction module concatenates continuous numerical values ​​of service life with low-dimensional embedding vectors of fault types to obtain a fused feature vector.

[0197] ;

[0198] The fully connected regression layer of the cost factor prediction module consists of two fully connected layers, employing the ReLU activation function to perform nonlinear mapping on the fused feature vector. The final output consists of three continuous values, corresponding to repair time, repair cost, and fault diagnosis difficulty, respectively. The fault diagnosis difficulty is the difficulty value for troubleshooting. The output formula for the fully connected layer is:

[0199] .

[0200] The input layer of the comprehensive cost fusion module consists of three neurons, corresponding to repair time, repair cost, and troubleshooting difficulty values, respectively. Input vector for cost factors:

[0201] ;

[0202] The hidden layer of the integrated cost fusion module consists of 1-2 fully connected layers using the ReLU activation function to capture the nonlinear coupling relationship between the three cost factors. The calculation formulas for the two hidden layers are as follows:

[0203] ;

[0204] The output layer of the comprehensive cost fusion module contains a neuron that uses a linear activation function to output a continuous comprehensive fault cost value. The output layer calculation formula is as follows:

[0205] ;

[0206] During training, the integrated cost fusion module uses the Mean Squared Error (MSE) loss function for training optimization. The MSE loss function is as follows:

[0207] ,

[0208] In the formula, The value of the mean squared error loss function for MSE is... The total number of samples, For the first The overall cost prediction value for each sample. For the first The actual comprehensive cost label value of each sample, it is worth noting, is... The technical serial numbers represented by are interchangeable across the various formulas in this article.

[0209] The ontology layer of the knowledge graph module defines a unified semantic framework and entity-relationship specification for the knowledge graph, specifically including:

[0210] Entity type definition: Divided into two categories: working condition entities and solution entities. Working condition entities include service life range and fault type, while solution entities include the number of maintenance personnel, whether replacement is required, and maintenance tools. Working condition entities are used for fault type matching, and solution entities are used for obtaining maintenance steps and strategies. Relationship type definition: Establishing the association between working condition entities and solution entities, and supplementing the hierarchical relationship between entities. Rule constraint definition: Setting knowledge rules based on engineering experience to ensure the rationality of maintenance decisions.

[0211] The data layer of the knowledge graph module, based on the ontology layer framework, completes the structured storage and vectorized fusion of historical maintenance experience data. Specifically, this includes: Structured mapping of historical data: transforming historical maintenance data into structured knowledge instances according to the ontology layer triplet specification, constructing a complete triplet set, and forming an instantiated knowledge graph; Vector embedding fusion: using the TransE knowledge graph embedding model, uniformly mapping ontology layer entities, relations, and data layer triplet instances to a d-dimensional latent space, obtaining entity embedding vectors and relation embedding vectors, achieving bidirectional mapping between vectors and the knowledge graph; the d-dimensionality mentioned above is a configurable latent space parameter.

[0212] The reasoning layer of the knowledge graph module: Based on the instantiated knowledge graph and vector embedding of the data layer, it completes the reasoning calculation from the input working condition vector to the output maintenance solution vector. Specifically, it includes: Input vector-graph entity matching: matching the input working condition vector with the graph entity. Encoded as latent space input vector Then, by vector similarity matching, the corresponding working condition entity subgraph is located, and relevant relationship vectors are extracted. ;

[0213] Translation vector aggregation generation: An attention-weighted aggregation method is used to aggregate the matched relation vectors and generate translation vectors;

[0214] Vector space reasoning and output generation: Performing core translation operations: ; Combining the validation results of ontology layer rules and constraints, decode The final maintenance solution vector is obtained. .

[0215] This embodiment also provides a method for constructing a knowledge graph module:

[0216] (a) Encode the input equipment condition vector and the output maintenance plan vector into numerical vectors respectively, and map them uniformly to the same d-dimensional latent space;

[0217] The equipment condition vector is composed of service life and fault type. The encoding method is as follows: service life is directly encoded with numerical values, and fault type is One-Hot encoded. The encoded result is the input vector.

[0218] The maintenance plan vector is: number of maintenance personnel, whether replacement is needed, and required tools. The encoding method is as follows: the number of maintenance personnel is directly encoded with numerical values, whether replacement is needed is encoded with binary values ​​(needed = 1, not needed = 0), and the required tools are encoded using One-Hot encoding. The encoded vector is then used to obtain the output vector.

[0219] The unified mapping is to the same The latent space is implemented using an input encoder and an output encoder. The input vector is encoded to obtain the latent space input vector, and the output vector is encoded to obtain the latent space output vector. To unify the latent space dimension, the input encoder and output encoder adopt an embedding layer structure;

[0220] (b) Construct a knowledge graph based on historical maintenance experience data, and use the TranS knowledge graph embedding model to learn the vector representations of entities and relations in the graph, and then aggregate them to obtain the translation vector corresponding to the current input vector. ;

[0221] The knowledge graph triples include entities and relations. Entities are divided into working condition entities and solution entities. Working condition entities include service life interval, fault location, and fault type. Solution entities include the number of maintenance personnel, whether replacement is required, and maintenance tools. Relationships are the correspondence between working condition entities and solution entities. Knowledge rules are also set. These knowledge rules are based on historical experience and are used to constrain the rationality of maintenance decisions.

[0222] The TransE knowledge graph embedding model embeds all entities and relations in the knowledge graph into the d-dimensional latent space of “(a)”, resulting in entity vectors and relation vectors.

[0223] The translation vector The aggregation method uses attention weighting, and the aggregation object is the vector with respect to the current input vector. Relational vectors in relevant knowledge graphs The physical meaning is the knowledge shift from equipment operating conditions to maintenance solutions; the longer the equipment's service life, the more severe the fault. The larger the vector magnitude;

[0224] (c) Step 3: In The vector translation operation is performed in the latent space to obtain the latent vector of the maintenance scheme, and then the maintenance decision result in the original output space is obtained by decoding.

[0225] The core formula for the vector translation operation is: ;

[0226] The decoder decodes the latent space output vector into the original output space maintenance plan vector. The decoded data, including the number of maintenance personnel, whether parts need to be replaced, and the required maintenance tools, are consistent with the maintenance plan format in historical experience data.

[0227] (d) Step 4: Supervised training of the entire decision-making model based on historical maintenance experience data, and incorporating rule constraints from the knowledge graph;

[0228] The objective function for the supervised training is: ,in , The first The equipment condition vector and the corresponding maintenance plan vector in the historical experience data. For the first The translation vector corresponding to each historical data point The model parameters are the loss function used to minimize the error between the translated latent space vector and the true latent space output vector. This indicates the search for an optimal set of parameters. , For the input encoder, For the output encoder, The square of the L2 norm;

[0229] The rules and constraints of the knowledge graph are used to penalize output results that do not conform to the knowledge rules, avoid unreasonable maintenance decisions, and ensure that the output maintenance solutions conform to historical experience and engineering practice.

[0230] As one embodiment, the fully mechanized mining support MBN network model includes an input layer, a hidden layer, a liquid state layer, and a readout layer connected in sequence.

[0231] The input layer of the MBN network model receives a normalized time-series data matrix. The hidden layer of the MBN network model consists of two layers of recursive restricted Boltzmann machines (rRBMs) stacked together to extract data features. The liquid state layer of the MBN network model fuses the extracted data features. The readout layer of the MBN network model is a lightweight linear layer.

[0232] As one embodiment, the output of the MBN network model may require additional processing. Dimensional and temporal features are extracted from the output of the fully mechanized mining support MBN network model to construct a comprehensive statistical feature vector. The dimensional features include: reconstruction error of each feature dimension, anomaly contribution of each dimension, number of anomalous feature dimensions, largest anomalous dimension number, and the combined feature of the top three anomalous dimensions. The temporal features include: anomaly initiation time, reconstruction error rise slope, error peak time, anomaly duration, error decay rate, and error sequence variance.

[0233] As one embodiment, the time-series data matrix should be preprocessed before being input into the MBN network model:

[0234] The long time series data is divided into several time series samples using the fixed time sliding window method, ensuring that the time step length and feature dimension of each sample are the set values, and invalid samples with insufficient time steps or / and missing more than 30% of feature dimensions are removed.

[0235] For missing values ​​in the feature matrix, linear interpolation, KNN interpolation, or temporal LSTM interpolation are used to fill them. The number of neighbors for KNN interpolation is 5 to 10, and the number of hidden layer neurons for temporal LSTM interpolation is 32 to 64. During the filling process, the trend features of the temporal data are strictly preserved to avoid destroying the temporal dependencies.

[0236] The Z-Score standardization method is used to standardize each feature dimension of the feature matrix separately. The standardization formula is as follows:

[0237] ,

[0238] In the formula, For the standardized feature dimensions, The feature dimensions before standardization. For the first The mean of each feature dimension, For the first The mean and standard deviation of each feature dimension.

[0239] As one embodiment, the construction process of the MBN network model includes:

[0240] (a) The input layer receives the preprocessed time series data matrix;

[0241] (b) The hidden layer consists of two stacked recursive restricted Boltzmann machines to obtain the temporal characteristics of the data; specifically:

[0242] ① Define the energy function of the first-layer rRBM:

[0243] ,

[0244] In the formula: superscript This represents the first layer rRBM. The visible layer state vector, This is the state vector of the first hidden layer. This is the parameter set for the first layer rRBM. and These are the bias vectors for the first visible layer and the first hidden layer, respectively. This is the weight matrix from the first visible layer to the first hidden layer. This is the recursive weight matrix within the first hidden layer;

[0245] ② Define the energy function of the second-layer rRBM:

[0246] ,

[0247] In the formula, the upper right superscript This indicates the second layer rRBM. This is the state vector of the first hidden layer. This is the parameter set for the second-layer rRBM. and These are the bias vectors for the first and second hidden layers, respectively. This is the weight matrix from the first hidden layer to the second hidden layer. This is the recursive weight matrix within the second hidden layer;

[0248] ③ Stack two layers of rRBM and define a joint energy function:

[0249] ,

[0250] In the formula, This is the parameter set for a two-layer rRBM stacked model;

[0251] ④ Calculate the state probability based on the joint energy function:

[0252] ,

[0253] In the formula, It is a natural constant. This is the partition function, used to normalize the probability distribution.

[0254] As one embodiment, during the initialization of the liquid state layer, i.e. At that time, the liquid state vector , .

[0255] As one embodiment, obtaining the fault type and fault occurrence probability based on the dimensional features and the temporal features includes constructing a comprehensive statistical feature vector based on the dimensional features and the temporal features, and processing the comprehensive statistical feature vector using a Bayesian classification model to obtain the fault type and the corresponding fault occurrence probability.

[0256] As one example, the Bayesian classification model calculation formula mentioned above is:

[0257] ,

[0258] In the formula, For comprehensive statistical feature vectors, For the first The posterior probability of each fault type For the first The characteristic conditional probability of each fault type For the first The prior probability of each fault type For the first Likelihood probability of each type of failure For the first The prior probability of each fault type To and Different counting numbers, To preset the total number of fault types, and They represent the first species and first Types of faults.

[0259] Example 3:

[0260] refer to Figure 3 This embodiment provides a computer device, including a processor and a memory connected to the processor. The memory stores a computer program. When the computer program is executed by the processor, it performs the steps of the maintenance strategy acquisition method for the fully mechanized mining support provided in Embodiment 1 or 2.

[0261] The computer device may be a server or an electronic terminal, as one embodiment, see reference. Figure 2 The computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The database stores data acquired and generated in the maintenance strategy acquisition method for fully mechanized mining supports. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the maintenance strategy acquisition method for fully mechanized mining supports provided in Embodiment 1 or 2.

[0262] Those skilled in the art will understand that Figure 2 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0263] The computer device provided in this embodiment has the same technical effects as that in Embodiment 1 or 2, and will not be described again here.

[0264] Example 4:

[0265] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the maintenance strategy acquisition method for the fully mechanized mining support provided in Embodiment 1 or Embodiment 2.

[0266] The computer-readable storage medium provided in this embodiment has the same technical effects as that in Embodiment 1 or 2, and will not be described again here.

[0267] Example 5:

[0268] This embodiment provides a computer program product storing a computer program that, when executed by a processor, implements the steps of the maintenance strategy acquisition method for the fully mechanized mining support provided in Embodiment 1 or Embodiment 2. The computer program product provided in this embodiment can be transmitted, distributed, and downloaded via the Internet in the form of signals.

[0269] The computer program product provided in this embodiment has the same technical effects as that in Embodiment 1 or 2, and will not be described again here.

[0270] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0271] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0272] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0273] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0274] In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0275] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0276] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for obtaining maintenance strategies for fully mechanized mining supports, characterized in that, include: Obtain the working parameter sequence of the fully mechanized mining support, and extract dimensional features and time-series features from the working parameter sequence; The fault type and the probability of fault occurrence are obtained based on the dimensional features and the temporal features. The maintenance cost and maintenance steps are determined based on the type of failure and the service life of the fully mechanized mining support. The maintenance steps are arranged according to the maintenance cost and the probability of failure to obtain a maintenance strategy.

2. The method for obtaining maintenance strategies for fully mechanized mining supports according to claim 1, characterized in that, The method for determining the maintenance cost and maintenance steps based on the fault type and the service life of the fully mechanized mining support includes: The maintenance time, maintenance cost, and troubleshooting difficulty value are obtained based on the fault type, the service life of the fully mechanized mining support, and the trained cost factor prediction network. The repair cost is determined by combining the repair time, repair cost, and troubleshooting difficulty. The maintenance steps are derived based on the service life of the fully mechanized mining support, the fault type, and a pre-constructed knowledge graph.

3. The method for obtaining maintenance strategies for fully mechanized mining supports according to claim 2, characterized in that, The process of obtaining maintenance time, maintenance cost, and troubleshooting difficulty values ​​based on the fault type, the service life of the fully mechanized mining support, and the trained cost factor prediction network includes: The trained cost factor prediction network includes an embedding layer, a feature splicing layer, and a fully connected regression layer; The fault type is mapped to a low-dimensional dense vector through the embedding layer, and the low-dimensional dense vector is obtained by the following formula: ， In the formula, It is a low-dimensional dense vector. This is the embedding layer weight matrix. For discrete category features, One-hot encoding for fault types, This is the embedding layer bias vector; The continuous values ​​of the service life of the fully mechanized mining support are extracted, and then concatenated with the low-dimensional dense vector through the feature splicing layer to obtain a fused feature vector; the fused feature vector is obtained by the following formula: ， In the formula, To fuse feature vectors, This represents a continuous numerical value of seniority. This is an operation that concatenates continuous numerical values ​​of years of service with a low-dimensional dense vector. The fully connected regression layer is used to perform a nonlinear mapping on the fused feature vector to obtain the repair time, the repair cost, and the troubleshooting difficulty value. ， In the formula, This is the output vector of the fully connected regression layer. For repair time, For repair costs, To check the difficulty value, This is the output layer weight matrix. This represents the number of fully connected layers in the fully connected regression layer. The first fully connected regression layer The weight matrix of each fully connected layer The first fully connected regression layer The bias vector of each fully connected layer The first fully connected regression layer The output of a fully connected layer This is the output layer bias vector. For the counting sequence number, The first fully connected regression layer The output of a fully connected layer The first fully connected regression layer The output of a fully connected layer The activation function is the activation operation. This represents the input to the fully connected regression layer.

4. The method for obtaining maintenance strategies for fully mechanized mining supports according to claim 2, characterized in that, The repair cost is calculated by integrating the repair time, repair cost, and troubleshooting difficulty, including: The repair time, the repair cost, and the troubleshooting difficulty value are input into the trained integrated cost fusion network, which includes an input layer, a hidden layer, and an output layer. The repair time, repair cost, and troubleshooting difficulty value are input into three neurons of the input layer to obtain the cost factor input vector; the cost factor input vector is obtained by the following formula: ， In the formula, This is the operation of inputting repair time, repair cost, and troubleshooting difficulty values ​​into the three neurons of the input layer, respectively. The cost factor input vector; The hidden layer is used to obtain the feature vector of the cost factor, which is obtained by the following formula: ， In the formula, The output of the first hidden layer of the integrated cost fusion network, The eigenvector of the cost factor, The activation function is the activation operation. This is the weight matrix of the first hidden layer of the integrated cost fusion network. This is the weight matrix of the second hidden layer of the integrated cost fusion network. The paranoia vector of the first hidden layer of the integrated cost fusion network. This is the paranoia vector of the second hidden layer of the integrated cost fusion network; The maintenance cost is obtained using the feature vectors of the output layer and the cost factor, and the maintenance cost is obtained by the following formula: ， In the formula, For the cost of repairs, This is the output layer bias vector. This is the output layer weight matrix.

5. The method for obtaining maintenance strategies for fully mechanized mining supports according to claim 2, characterized in that, The maintenance steps, derived from the age of the fully mechanized mining support, the fault type, and a pre-constructed knowledge graph, include: The service life of the fully mechanized mining support and the fault type are concatenated into an input condition vector; the input condition vector is encoded into a latent space input vector. The input working condition vector is matched for vector similarity in the pre-constructed knowledge graph to obtain a relevant relation vector. Aggregate the relevant relation vectors to obtain the translation vector; Based on the translation vector and the latent space input vector, a maintenance scheme vector is obtained; the maintenance steps are obtained by decoding the maintenance scheme vector; the maintenance scheme vector is obtained by the following formula: ， In the formula, For the input working condition vector, This is an operation that concatenates the service life and fault type of the fully mechanized mining support into a vector. It is a translation vector. This is a correlation vector. This is a vector aggregation operation. This is used to introduce the subsequent filtering conditions. It is a set of related relation vectors. For maintenance solution vectors, The input vector is the latent space vector. This is the decoded repair scheme vector. This represents the maintenance steps extracted from the decoded maintenance plan vector.

6. The method for obtaining maintenance strategies for fully mechanized mining supports according to claim 1, characterized in that, The extraction of dimensional features and temporal features from the working parameter sequence includes: The working parameter sequence is sequentially subjected to dimension alignment, missing value imputation, and feature dimension standardization to obtain the preprocessed working parameter sequence. Obtain a trained MBN network model, which includes a hidden layer, a liquid state layer, and a readout layer; input the preprocessed working parameter sequence into the hidden layer of the MBN network model to obtain a data temporal feature matrix; The liquid state matrix is ​​updated step-by-step according to the data time series feature matrix to obtain liquid state vectors at different time steps; the liquid state vectors at each time step are concatenated to obtain the liquid state matrix; the liquid state matrix is ​​as follows: ， In the formula, For time steps, To update to the Liquid state vector at each time step. It is a time-varying activation function. The first of the time series feature matrix of the data The eigenvectors of the row, The data temporal feature matrix is ​​mapped from the hidden layer to a fixed sparse weight matrix in the liquid state layer. To update to the Liquid state vector at each time step. For the fixed sparse recursive weight matrix of the liquid state layer, A fixed bias vector is applied to the liquid state layer. For the number of time steps, The liquid state matrix, To update to the The liquid state vector at each time step, with the upper right subscript... To represent the transpose of a matrix, Represent real numbers, This represents the number of liquid neurons in the liquid state layer. The liquid state vector in the liquid state matrix is ​​reconstructed step-by-step using the readout layer to obtain the reconstructed feature vector. The reconstructed time-series data feature matrix is ​​then obtained based on the reconstructed feature vector. The formula for obtaining the reconstructed time-series data feature matrix is ​​as follows: ， In the formula, For the corresponding number Reconstructed feature vectors at each time step This is the output layer weight matrix. This is the output layer bias vector. To reconstruct the feature matrix of time series data; Based on the reconstructed time-series data feature matrix, dimensional features and time-series features are obtained.

7. The method for obtaining maintenance strategies for fully mechanized mining supports according to claim 6, characterized in that, The step of obtaining dimensional features and time-series features based on the reconstructed time-series data feature matrix includes: The feature dimension reconstruction error is obtained based on the reconstructed time-series data feature matrix and the data time-series feature matrix. The feature dimension reconstruction error is obtained by the following formula: , In the formula, For the first Reconstruction error in each feature dimension For the counting sequence number, The first of the time series feature matrix of the data The time step and the first Feature values ​​of each feature dimension To reconstruct the first feature matrix of time series data The time step and the first Feature values ​​of each feature dimension; The dimension anomaly contribution is obtained based on the feature dimension reconstruction error, and the dimension anomaly contribution is obtained by the following formula: ， In the formula, For the first The contribution of dimensional anomalies in each feature dimension; Obtain the feature dimension error threshold, and count the number of feature dimension reconstruction errors that are greater than the feature dimension error threshold to obtain the number of abnormal feature dimensions. Traverse all feature dimension reconstruction errors, find the number corresponding to the largest feature dimension reconstruction error, and obtain the number of the largest outlier dimension. Sort all feature dimension reconstruction errors, extract the top three feature dimension reconstruction errors with the largest errors from the sequence, and form a three-dimensional anomaly dimension combination feature. The feature dimension reconstruction error, the dimension anomaly contribution, the number of anomalous feature dimensions, the maximum anomalous dimension number, and the three-dimensional anomalous dimension combination feature are used as the dimension class feature; The time step error sequence is obtained based on the reconstructed time series data feature matrix and the data time series feature matrix. The formula for obtaining the time step error sequence is as follows: ， In the formula, The time step error sequence, The number of feature dimensions; Obtain the time step error threshold, and find the starting time step from the time step error sequence that satisfies the following condition: starting from the starting time step, the time step error sequence corresponding to 3 to 5 consecutive time steps first appears to be greater than the time step error threshold; the starting time step is taken as the abnormal start time. Extract the time step corresponding to the maximum error value from the time step error sequence to obtain the error peak time; The slope of the reconstruction error is obtained based on the anomaly start time and the error peak time. The formula for obtaining the slope of the reconstruction error is as follows: ， This is it. To reconstruct the slope of the error rise, This represents the error at the peak time in the time-step error sequence. This represents the error at the abnormal starting point within the peak error time. This is the time when the error peaks. This is the time when the anomaly began; The duration of the anomaly is obtained by detecting the time difference from the start time of the anomaly to the current detection time step. The error decay rate is obtained based on the peak error time and the time step error sequence. The formula for obtaining the error decay rate is as follows: , In the formula, For the error decay rate, This represents the error at the abnormal end time or the current detection time step. This refers to the end time of the anomaly or the current detection time step. The variance of the error sequence fluctuation is obtained based on the duration of the anomaly and the time step error sequence. The formula for obtaining the variance of the error sequence fluctuation is as follows: ， In the formula, The variance of the error sequence fluctuation. The duration of the anomaly. This represents the average error value within the duration of the anomaly in the time-step error sequence. This is the abnormal termination time; The anomaly initiation time, the reconstruction error rise slope, the error peak time, the anomaly duration, the error decay rate, and the error sequence variance are used as the time-series features. The step of obtaining the fault type and fault occurrence probability based on the dimensional class features and the temporal class features includes: A comprehensive statistical feature vector is constructed based on the dimensional features and the temporal features; A Bayesian classification model is used to process the comprehensive statistical feature vector to obtain the posterior probability of each fault type, and this posterior probability is used as the probability of fault occurrence. The formula for obtaining the posterior probability is as follows: ， In the formula, For comprehensive statistical feature vectors, For the first The posterior probability of each fault type For the first The characteristic conditional probability of each fault type For the first The prior probability of each fault type For the first Likelihood probability of each type of failure For the first The prior probability of each fault type To and Different counting numbers, To preset the total number of fault types, and They represent the first species and first Types of faults.

8. A computer device, characterized in that, It includes a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it performs the steps of the maintenance strategy acquisition method for the fully mechanized mining support as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the steps of the maintenance strategy acquisition method for the fully mechanized mining support as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the maintenance strategy acquisition method for the fully mechanized mining support as described in any one of claims 1 to 7.

Images