Intelligent mine state adjusting system and method based on Internet of Things

Abstract

The invention discloses a mine state intelligent adjusting system and method based on the Internet of Things, and the system comprises a state extraction module which obtains an initial state factor from an operation parameter, and carries out the deep clustering optimization of the initial state factor, and extracts a state factor; the deep application module is used for carrying out global modeling on the state factors based on a Transform architecture to obtain global representation of the state factors, and analyzing importance sequence representation of the state factors under different time scales based on the global representation; and the decision control module is used for selecting an optimal time interval for the state parameters corresponding to the state factors in each sample according to the importance sequence representation to carry out data acquisition, acquiring the state parameters acquired at each time interval to observe, immediately sending an alarm and adjusting the equipment operation parameters when an abnormal condition occurs, and controlling the operation of the equipment. And the importance of the acquired data is not influenced by the difference of change rates of different equipment states.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to an intelligent mine state adjustment system and method based on the Internet of Things. Background Art

[0002] The underground working environment is complex, with many safety hazards such as gas explosion, water seepage, and roof collapse. Traditional safety monitoring methods have problems such as limited monitoring range and untimely information transmission, making it difficult to detect and warn of potential dangers in a timely manner. For example, in coal mines, the accident rate of rubber-tyred trackless vehicles has been on the rise in recent years, which has a great impact on underground production safety, causing personal injury and economic losses. Internet of Things technology can realize real-time and comprehensive monitoring of mine equipment and environment, detect abnormalities in a timely manner and issue warnings, and provide strong guarantees for safe production.

[0003] At present, in traditional mine state adjustment systems, fixed time intervals are usually used to sample and input equipment state data. This fixed time interval method does not take into account the differences in the change rates of different equipment states. For example, the current parameters of mine equipment will fluctuate greatly in an instant when the equipment is started, the load changes, etc., while the temperature parameters of the equipment change slowly over a long period of time. If a fixed time interval is used uniformly, for fast-changing parameters such as current, key change information may be missed due to the large time interval. For slowly changing parameters such as temperature, the long-term change trend cannot be accurately captured due to the short collection time interval, and the true operating status of the equipment cannot be accurately obtained, which affects the control decision of the mine equipment. Therefore, an intelligent mine state adjustment system and method based on the Internet of Things is proposed to solve the above problems. Summary of the invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned purpose, the present invention proposes the following technical solutions: An intelligent mine state adjustment system based on the Internet of Things, comprising: Data acquisition module: collects mine equipment data based on the Internet of Things technology and extracts operating parameters from the mine equipment data; State extraction module: obtains preliminary state factors from operating parameters, and then performs deep clustering optimization on the preliminary state factors to extract state factors; Deep Application Module: Globally model the state factors based on the Transformer architecture, obtain the global representation of the state factors, and analyze the importance sequence representation of the state factors at different time scales based on the global representation; Decision control module: According to the importance sequence representation, the optimal time interval is selected for data collection for the state parameters corresponding to the state factors in each sample, and the state parameters collected at each time interval are obtained for observation. When an abnormal situation occurs, an alarm is immediately sent and the equipment operation parameters are adjusted.

[0005] The operating parameters include fast-changing operating parameters and slow-changing operating parameters; The fast-changing operating parameters include vibration parameters and voltage parameters; The slowly varying operating parameters include temperature parameters and pressure parameters.

[0006] The preliminary state factor is collected in the following manner: Based on the operation parameters, feature data is extracted through a convolutional neural network to construct a conditional generative adversarial network. , to operate on the category labels of the parameter data For conditions; An integrated model consisting of multiple autoencoders with different structures is established. Each autoencoder extracts and compresses features of the original data or generated data from different angles. The encoding results of multiple autoencoders are fused, and the fused features are represented as preliminary state factors. .

[0007] The integrated model consists of a sparse autoencoder and a variational autoencoder; The category label By setting normal threshold ranges for various operating parameters according to the design standards of mine equipment, different operating states of different operating parameters are obtained to obtain category labels. .

[0008] The state factor extraction process is: Initial state factor Conduct further deep clustering optimization; Extracting preliminary state factors through deep learning network Based on the high-level feature representation, a series of clustering variables are obtained, and the category labels are As an evaluation indicator of the state factor, through the objective function Implement clustering; Assume the state factor is expressed as , state factor The information gain is , get the sum of state factor importance ; By sum of state factor importance Constructing the objective function , expressed as:

[0009] in, represents the balance coefficient, is the clustering variable; By minimizing the objective function, the initial state factor Perform clustering to obtain the state factor .

[0010] The clustering variables include: The compactness measure of clustering is , measures the closeness of high-level feature representation within the same cluster; The separation measure is , measures the degree of separation of high-level feature representations between different clusters; The label consistency metric is , measure the clustering results and class labels degree of match.

[0011] The process of obtaining the global representation of the state factor is: The dependencies between state factors are captured through the self-attention mechanism, and the features are gradually refined with the help of a multi-layer structure; The state factor Each state factor is mapped to a vector, and the position code is added to reflect the order information to obtain the input sequence ; State Factor Based on the input sequence Sequentially input into a model constructed by stacking multiple Transformer blocks in sequence; In a single self-attention mechanism stage, for each input sequence corresponding to the state factor , get the attention weight ; The query vector, key vector and value vector are obtained through linear transformation, and the attention weights are calculated. Perform weighted summation on the value vector to get the attention output of a single head ; Then output multiple single-head attention Splice and linearly transform to get the output result , the output Input into the feedforward neural network to transform and refine the features ; By stacking multiple Transformer blocks, Perform feature extraction and integration to obtain a global representation of the state factor .

[0012] The acquisition process of the importance sequence representation is: Based on global representation , add a fully connected layer; Suppose the weight matrix of the added fully connected layer is , the bias vector is ; By formula Get an output value, and then input the output value into the activation function In the activation function Map the output value to the optimal range, which is the importance sequence representation .

[0013] The process of obtaining the optimal time interval is as follows: Based on importance sequence , divided into categories and sort them from high to low; For each importance sequence Corresponding state factor, calculate the rate of change ; Assume that importance is divided into three categories: , set a change rate threshold ; Based on the rate of change and change rate threshold Get the best time interval.

[0014] A method for intelligently adjusting a mine state based on the Internet of Things, comprising: S1: Collect mine equipment data based on the Internet of Things technology and extract operating parameters from the mine equipment data; S2: Obtain preliminary state factors from the operating parameters, and then perform deep clustering optimization on the preliminary state factors to extract state factors; S3: Globally model the state factor based on the Transformer architecture, obtain the global representation of the state factor, and analyze the importance sequence representation of the state factor at different time scales based on the global representation; S4: According to the importance sequence representation, select the best time interval for data collection for the state parameters corresponding to the state factors in each sample, and obtain the state parameters collected at each time interval for observation. When an abnormal situation occurs, send an alarm immediately and adjust the equipment operation parameters.

[0015] The present invention has the following beneficial effects: In the present invention, firstly, operating parameters are extracted from a large amount of collected data, so that data processing is more targeted. The operation of mine equipment will generate a large amount of data. By accurately extracting operating parameters, attention can be focused on factors that have a direct impact on the operating status of the equipment, reducing unnecessary data interference and improving the efficiency and effectiveness of data analysis. Secondly, preliminary status factors are obtained from the operating parameters, which completes the preliminary screening and integration of the original data, converts the operating parameters into status factors with more analytical value, and lays the foundation for subsequent in-depth analysis. This helps to more accurately characterize the operating status of the equipment and improve the ability to identify abnormal status of the equipment; Then, the state factors are globally modeled based on the Transformer architecture, which can fully capture the mutual relationship and dependency information of the state factors at different time scales and spatial dimensions, and obtain the global representation of the state factors. This enables the model to understand the operating status of the equipment from a macro perspective and avoid misjudgment caused by local information analysis. Based on the global representation, the importance sequence representation of the state factors at different time scales can be analyzed to clarify the importance and role of each state factor in the operation of the equipment. This helps to focus resources and attention on key state parameters in equipment monitoring and maintenance, give priority to factors that have a greater impact on equipment operation, and improve the accuracy and effectiveness of equipment management. Finally, the optimal collection time interval is selected according to the importance of the state factors to realize the intelligent management of data collection, which can collect key and important factors at a high frequency and reduce the collection frequency of non-key factors, so as to make better targeted control decisions for mine equipment. The importance of collected data will not be affected by the difference in the rate of change of different equipment states. When an abnormal situation occurs, the equipment operating parameters can be adjusted immediately, realizing real-time monitoring and automatic control of mine equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a system block diagram of a mine status intelligent adjustment system and method based on the Internet of Things proposed by the present invention.

[0017] Figure 2 This is a method step diagram of a mine status intelligent adjustment system and method based on the Internet of Things proposed by the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, 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.

[0019] Embodiment 1: Figure 1 As shown, the present invention proposes an intelligent mine state adjustment system based on the Internet of Things, comprising: Data acquisition module: collects mine equipment data based on the Internet of Things technology and extracts operating parameters from the mine equipment data; The operating parameters include fast-changing operating parameters and slow-changing operating parameters; The fast-changing operating parameters include vibration parameters and voltage parameters, and the slow-changing operating parameters include temperature parameters and pressure parameters; Specifically, the slowly varying parameter collection method is: Under normal conditions, the temperature sensor collects data every 10 seconds. When the temperature change rate is detected to exceed 0.2°C per minute, the collection frequency is automatically increased to once every 2 seconds. When the pressure sensor is normal, data is collected every 15 seconds. When the pressure change rate exceeds 0.2MPa per minute, the collection frequency is adjusted to once every 3 seconds. The collected data include temperature values, pressure values ​​and corresponding timestamps; Fast-changing parameter collection method: The initial acquisition frequency of the vibration sensor is set to When the equipment is in startup, braking, overload or other abnormal working conditions, the built-in intelligent algorithm monitors the change rate of characteristic parameters such as the peak value of the vibration signal in real time. When the change rate exceeds the preset threshold, the acquisition frequency can be increased to ; The initial acquisition frequency of the current and voltage sensors is , when the equipment operating status changes, such as the current fluctuation exceeds the rated current Or the voltage fluctuation exceeds the rated voltage When ; The collected data include vibration data, current value, voltage value and corresponding timestamp.

[0020] State extraction module: obtains preliminary state factors from operating parameters, and then performs deep clustering optimization on the preliminary state factors to extract state factors; The initial state factor extraction process is: Based on the operation parameters, feature data is extracted through a convolutional neural network to construct a conditional generative adversarial network. , to operate on the category labels of the parameter data (labels corresponding to different operating states of the equipment) as conditions; Among them, conditional generative adversarial network The goal of the generator is to learn to generate data samples that are similar to the real operating parameter data features but have diversity , is a random noise vector; Conditional Generative Adversarial Networks The discriminator is used to distinguish the real data and generate data ; Specifically, the category label is obtained as follows: According to the design standards of the mine equipment, set normal threshold ranges for various operating parameters, such as: For the current parameters of the underground motor, the current fluctuates within a certain range during normal operation. When the current exceeds the normal threshold, it is marked as "overload operation state"; Below the normal threshold, it is marked as "light load operation state"; If it is within the normal range, it is marked as "normal operation status"; For temperature parameters, if the temperature of key components of the equipment exceeds the safety threshold, it is marked as "abnormal high temperature state", and if it is within the safety range, it is marked as "normal temperature state"; Obtain different operating states of these different operating parameters, i.e., category labels; An integrated model consisting of multiple (5) autoencoders with different structures is established. Each autoencoder extracts and compresses the original data or generated data from different angles. The encoding results of multiple autoencoders are fused, and the fused features are represented as preliminary state factors. , where i is the index of the number of initial state factors; The state factor extraction process is: Specifically, the integrated model consists of a sparse autoencoder and a variational autoencoder; Initial state factor Conduct further deep clustering optimization; Assume the state factor is expressed as ; The high-level feature representation of the initial state factor is extracted through the deep learning network, and a series of clustering variables are obtained based on the high-level feature representation. As an evaluation indicator of the state factor, through the objective function Implement clustering; Specifically, a series of clustering variables include: The compactness measure of clustering is , used to measure the compactness of high-level feature representations within the same cluster; The separation measure is , measures the degree of separation of high-level feature representations between different clusters; The label consistency metric is , measure the clustering results and class labels The degree of matching; Obtain the importance of the state factor by calculating the information gain index. Set the state factor The information gain is , then the sum of the importance of state factors is: , due to the state factor The information gain is based on the initial state factor Get, so the sum of the state factor importance contains elements, and the initial state factor The quantity is the same; Then based on the sum of state factor importance The objective function is constructed as: ,in, represents the balance coefficient; By minimizing the objective function, the preliminary state factors are clustered to obtain the final state factors. ,in, is the number index of the state factors, which is the same as the number of the preliminary state factors. One preliminary state factor corresponds to one state factor after clustering optimization. Specifically, in the clustering optimization process, the objective function The clustering division and the screening of state factors will be continuously adjusted. When the objective function converges to a certain degree or meets the preset stop condition, the clustering result and the corresponding parameters are the optimized state factors.

[0021] Deep Application Module: Globally model the state factors based on the Transformer architecture, obtain the global representation of the state factors, and analyze the importance sequence representation of the state factors at different time scales based on the global representation; Based on the Transformer architecture, the state factor is globally modeled and the process of obtaining the global representation of the state factor is as follows: The dependencies between state factors are captured through the self-attention mechanism, and the features are gradually refined with the help of a multi-layer structure; The state factor Each state factor is mapped to a vector, and the position code is added to reflect the order information to obtain the input sequence , where the number of input sequences is the same as the number of state factors. ; State Factor Based on the input sequence Sequentially input into a model constructed by stacking multiple (12) Transformer blocks in sequence; The core components of each Transformer block are composed of a multi-head self-attention mechanism and a feed-forward neural network, that is, one Transformer corresponds to one head of the multi-head self-attention mechanism; In a single self-attention mechanism stage, for each input sequence corresponding to the state factor , get the attention weight ; The query vector, key vector and value vector are obtained through linear transformation, and the attention weights are calculated. Perform weighted summation on the value vector to get the attention output of a single head ; Then multiple (12) single-head attention outputs Concatenate and linearly transform the results to obtain the output result. , the output Input into the feedforward neural network to further transform and refine the features. The formula is expressed as:

[0022] in, represents a feed-forward neural network, express Activation function, is the weight matrix of the first layer of the feedforward neural network, with a dimension of , is the weight matrix of the second layer of the feedforward neural network, and its dimension is , is the bias vector of the first layer of the feedforward neural network, and its dimension is , is the bias vector of the second layer of the feedforward neural network, and its dimension is ; Furthermore, represents the model dimension, that is, the dimension of the feature vector of the data input to the feedforward neural network (the output of the multi-head self-attention mechanism), represents the dimension of the intermediate layer of the feedforward neural network, and ; Then, by stacking multiple Transformer blocks, Perform feature extraction and integration to finally obtain a global representation of the state factor Similarly, a state factor corresponds to a global representation of a state factor, so the number is ; Specifically, the output of the previous Transformer block is used as the input of the next Transformer block. After each Transformer block, the features of the state factors are further extracted and integrated. For example, the output of the first Transformer block is normalized and residually connected, and then used as the input of the second Transformer block. The calculation process of the multi-head self-attention mechanism and the feedforward neural network is repeated to obtain , and so on. As the Transformer blocks are stacked, the model can continuously optimize the relationship between state factors and finally obtain a global representation of the state factors. ; Based on global representation The process of analyzing the importance sequence representation of state factors at different time scales is: Getting a global representation Finally, a fully connected layer is added to analyze the importance of the state factor; Specifically, due to the global representation The comprehensive information of all state factors at different time scales is integrated, and the importance of state factors is analyzed from this global representation. In the analysis, the importance of each state factor at different time scales is obtained and presented in the form of a sequence, which is convenient for subsequent targeted processing according to the importance of the state factor, such as control strategy formulation, fault warning, etc.; Suppose the weight matrix of the added fully connected layer is , the bias vector is ; Among them, the fully connected layer weight matrix The dimension is ,in, is the output dimension of the fully connected layer, and (number of state factors, so as to obtain the importance score corresponding to each state factor), The dimension is ; By formula Get an output value, and then input the output value into the activation function In the activation function Map the output value to the optimal range, which is the importance sequence representation , a global representation corresponds to an importance sequence representation, so the number is ; Specifically, the sum of all importance sequences output through the above steps is 1. It is The proportion of a state factor in the importance of all state factors. The higher the proportion, the more important the state factor is.

[0023] Decision-making control module: According to the importance sequence representation, the optimal time interval is selected for data collection for the state parameters corresponding to the state factors in each sample, and the state parameters collected at each time interval are obtained for observation. When an abnormal situation occurs, an alarm is immediately sent and the equipment operation parameters are adjusted; The process of obtaining the optimal time interval is: Based on importance sequence , divided into categories, and sort them from high to low, expressed as , where the importance is ranked as , each category contains several importance sequence values, and the total number of all importance sequences is ; For each importance sequence The corresponding state factor is calculated to determine its rate of change over a period of time. The calculation process is:

[0024] in, For the time step The state factor value at For the time step The state factor value at time , and the state factor change rate ; The rate of change of state factors for different categories The corresponding devices and their change rate ranges are used to establish the mapping rules of time intervals. The importance is divided into three categories, namely , and set a change rate threshold , based on the rate of change and change rate threshold Get the optimal time interval, specifically: For the importance sequence (High) and the rate of change Greater than the change rate threshold The device corresponding to the status factor selects the first time interval (High importance and high rate of change, with a minimum time interval of 5 seconds to 1 minute) for data collection; For the importance of (High) and the rate of change Less than or equal to the change rate threshold The device corresponding to the status factor selects the second time interval (High importance and low rate of change, with shorter time intervals of 1 to 5 minutes) for data collection; For the importance of (Medium) and the rate of change Greater than the change rate threshold The device corresponding to the status factor selects the third time interval (Medium importance and high rate of change, use shorter time intervals of 5 to 10 minutes) for data collection; For the importance of (Medium) and the rate of change Less than or equal to the change rate threshold The device corresponding to the status factor selects the fourth time interval (medium importance and low rate of change, use longer time intervals of 10 to 15 minutes) for data collection; For the importance of (low) state factor corresponding to the device, regardless of the rate of change How to choose the fifth time interval (low importance and arbitrary change rate, long time interval of 20 minutes) for data collection; Specifically, the importance of state factors is divided into three categories ( high, middle, This classification method helps to distinguish the degree of influence of different state factors on the operation of equipment or system. For example, in the power system, the voltage, current and other parameters of key equipment can be classified as The operating status parameters of some auxiliary equipment can be classified into The parameters with less influence on system operation by equipment surface temperature can be classified as ; According to the above mapping rules, for the equipment corresponding to each state factor in each sample, the corresponding collection time interval is determined in combination with its importance category and change rate. For the equipment corresponding to the state factor with high importance and fast change, a shorter time interval is selected for high-frequency collection. For the equipment corresponding to the state factor with low importance and slow change, a longer time interval is selected. The parameter data is monitored for abnormal conditions according to the time interval, and the equipment is controlled according to the monitoring results of the data. Formulate corresponding equipment control strategies according to the abnormal conditions of different status factors and their impact on equipment operation; For example, when the voltage, current and other parameters of key equipment ( When a parameter of an auxiliary device ( When the equipment surface temperature ( When the status factor of the class shows an upward trend but does not affect the overall operation, an early warning is issued and recorded, and its changing trend is continuously monitored.

[0025] Embodiment 2: Figure 2 As shown, a method for intelligently adjusting the state of a mine based on the Internet of Things includes: S1: Collect mine equipment data based on the Internet of Things technology and extract operating parameters from the mine equipment data; S2: Obtain preliminary state factors from the operating parameters, and then perform deep clustering optimization on the preliminary state factors to extract state factors; S3: Globally model the state factor based on the Transformer architecture, obtain the global representation of the state factor, and analyze the importance sequence representation of the state factor at different time scales based on the global representation; S4: According to the importance sequence representation, select the best time interval for data collection for the state parameters corresponding to the state factors in each sample, and obtain the state parameters collected at each time interval for observation. When an abnormal situation occurs, send an alarm immediately and adjust the equipment operation parameters.

[0026] In the application, several formulas involved are calculated by removing dimensions and taking their numerical values. The formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent actual situation. Some coefficients or weights in the formulas are set by technicians in this field according to actual conditions, so they will not be elaborated here.

[0027] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. A person of ordinary skill in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented in electronic hardware, or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0028] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent mine status adjustment system based on the Internet of Things, characterized in that: include: Data acquisition module: collects mine equipment data based on the Internet of Things technology and extracts operating parameters from the mine equipment data; State extraction module: obtains preliminary state factors from operating parameters, and then performs deep clustering optimization on the preliminary state factors to extract state factors; Deep Application Module: Globally model the state factors based on the Transformer architecture, obtain the global representation of the state factors, and analyze the importance sequence representation of the state factors at different time scales based on the global representation; Decision control module: According to the importance sequence representation, the optimal time interval is selected for data collection for the state parameters corresponding to the state factors in each sample, and the state parameters collected at each time interval are obtained for observation. When an abnormal situation occurs, an alarm is immediately sent and the equipment operation parameters are adjusted.

2. According to the Internet of Things-based intelligent mine status adjustment system of claim 1, it is characterized in that: The operating parameters include fast-changing operating parameters and slow-changing operating parameters; The fast-changing operating parameters include vibration parameters and voltage parameters; The slowly varying operating parameters include temperature parameters and pressure parameters.

3. The intelligent mine state adjustment system based on the Internet of Things according to claim 1 is characterized in that: The preliminary state factor is collected in the following manner: Based on the operation parameters, feature data is extracted through a convolutional neural network to construct a conditional generative adversarial network. , to operate on the category labels of the parameter data For conditions; An integrated model consisting of multiple autoencoders with different structures is established. Each autoencoder extracts and compresses features of the original data or generated data from different angles. The encoding results of multiple autoencoders are fused, and the fused features are represented as preliminary state factors. .

4. The intelligent mine state adjustment system based on the Internet of Things according to claim 3 is characterized in that: The integrated model consists of a sparse autoencoder and a variational autoencoder; The category label By setting normal threshold ranges for various operating parameters according to the design standards of mine equipment, different operating states of different operating parameters are obtained to obtain category labels. .

5. The intelligent mine state adjustment system based on the Internet of Things according to claim 1 is characterized in that: The state factor extraction process is: Initial state factor Conduct further deep clustering optimization; Extracting preliminary state factors through deep learning network Based on the high-level feature representation, a series of clustering variables are obtained, and the category labels are As an evaluation indicator of the state factor, through the objective function Implement clustering; Assume the state factor is expressed as , state factor The information gain is , get the sum of state factor importance ; By sum of state factor importance Constructing the objective function , expressed as: ； in, represents the balance coefficient, is the clustering variable; By minimizing the objective function, the initial state factor Perform clustering to obtain the state factor .

6. The intelligent mine status adjustment system based on the Internet of Things according to claim 5 is characterized in that: The clustering variables include: The compactness measure of clustering is , measures the closeness of high-level feature representation within the same cluster; The separation measure is , measures the degree of separation of high-level feature representations between different clusters; The label consistency metric is , measure the clustering results and class labels degree of match.

7. The intelligent mine status adjustment system based on the Internet of Things according to claim 1 is characterized in that: The process of obtaining the global representation of the state factor is: The self-attention mechanism is used to capture the dependencies between state factors, and the multi-layer structure is used to gradually refine the features. Each state factor is mapped to a vector, and the position code is added to reflect the order information to obtain the input sequence ; State Factor Based on the input sequence The input sequence is sequentially fed into a model constructed by stacking multiple Transformer blocks in sequence. In a single self-attention mechanism stage, the state factor corresponding to each input sequence is , get the attention weight ; The query vector, key vector and value vector are obtained through linear transformation, and the attention weights are calculated. Perform weighted summation on the value vector to get the attention output of a single head ; Then output multiple single-head attention Splice and linearly transform to get the output result , the output Input into the feedforward neural network to transform and refine the features ; By stacking multiple Transformer blocks, Perform feature extraction and integration to obtain a global representation of the state factor .

8. The intelligent mine status adjustment system based on the Internet of Things according to claim 1 is characterized in that: The acquisition process of the importance sequence representation is: Based on global representation , add a fully connected layer; Suppose the weight matrix of the added fully connected layer is , the bias vector is ; By formula Get an output value, and then input the output value into the activation function In the activation function Map the output value to the optimal range, which is the importance sequence representation .

9. The intelligent mine state adjustment system based on the Internet of Things according to claim 1 is characterized in that: The process of obtaining the optimal time interval is as follows: Based on importance sequence , divided into categories and sort them from high to low; For each importance sequence Corresponding state factor, calculate the rate of change ; Assume that importance is divided into three categories: , set a change rate threshold ; Based on the rate of change and change rate threshold Get the best time interval.

10. A method for intelligently adjusting the state of a mine based on the Internet of Things, using any system described in claims 1 to 9, characterized in that: include: S1: Collect mine equipment data based on the Internet of Things technology and extract operating parameters from the mine equipment data; S2: Obtain preliminary state factors from the operating parameters, and then perform deep clustering optimization on the preliminary state factors to extract state factors; S3: Globally model the state factor based on the Transformer architecture, obtain the global representation of the state factor, and analyze the importance sequence representation of the state factor at different time scales based on the global representation; S4: According to the importance sequence representation, select the best time interval for data collection for the state parameters corresponding to the state factors in each sample, and obtain the state parameters collected at each time interval for observation. When an abnormal situation occurs, send an alarm immediately and adjust the equipment operation parameters.

Images