Equipment performance prediction method based on coal mine vertical field large model

Abstract

The invention provides an equipment performance prediction method based on a coal mine vertical field large model. Firstly, historical operation data of coal mine equipment are collected and combined with knowledge of a coal mine vertical field to construct a field knowledge graph; then constructing an equipment performance prediction preset large model, organizing and understanding modeling contents by using a domain knowledge graph, performing fine tuning by using a deep learning method, and performing supervised learning by using a plurality of label data to obtain an equipment performance prediction large model; and then real-time operation data of coal mine equipment is collected and input into the equipment performance prediction large model for performance prediction, and when the model predicts a potential fault, an alarm is triggered and a maintenance suggestion is provided. Based on a vertical field large model, the accuracy and sensitivity of prediction are improved for specific environments and requirements of the coal mining industry; meanwhile, an alarm and a maintenance suggestion can be given out immediately when the equipment shows an initial fault sign, so that problems can be solved in time, and fault extension is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mine equipment performance prediction and early warning, and in particular to an equipment performance prediction method based on a large model of a coal mine vertical field. Background Art

[0002] Effective monitoring of key mechanical equipment in mines is an important factor in ensuring production safety and efficiency. Traditional equipment monitoring methods mainly rely on regular manual inspections, which are not only time-consuming and costly, but may also miss important early fault signals due to untimely inspections. In addition, manual monitoring is difficult to fully cover the operating status of all equipment, especially when the mine environment conditions change rapidly, it is easy to miss key anomalies.

[0003] Existing automatic monitoring systems and methods usually rely only on basic sensor information or simple data analysis, which limits their accuracy and adaptability. These systems often lack the ability of deep learning and predictive maintenance, and cannot provide dynamic solutions for specific equipment problems, resulting in insufficient response or accuracy when dealing with complex or unexpected problems. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting equipment performance based on a large model of a vertical field of a coal mine, so as to solve the deficiencies of the above-mentioned prior art.

[0005] To achieve the above object, the present invention adopts the following technical solution:

[0006] A method for predicting equipment performance based on a large model of a coal mine vertical field comprises the following steps:

[0007] S1. Collect historical operation data of coal mine equipment and pre-process the collected data;

[0008] S2, combining the pre-processed historical operation data of coal mine equipment with the knowledge in the vertical field of coal mines to construct a domain knowledge graph;

[0009] S3. Based on the general large model, the domain knowledge graph is combined to train the model, and the deep learning method is used for fine-tuning. A number of labeled data are used for supervised learning to obtain a large model for device performance prediction and deploy it in practical applications.

[0010] S4, collect real-time operation data of coal mine equipment and perform preprocessing;

[0011] S5. Input the preprocessed real-time operation data of coal mine equipment into the equipment performance prediction model for performance prediction. When the model predicts a potential failure, it triggers an alarm and provides maintenance suggestions.

[0012] Furthermore, the historical operation data of the coal mine equipment is obtained from the equipment monitoring system, and the historical operation data of the mining electromechanical equipment includes reliability data, maintenance records, fault types and frequencies or other data that can characterize the historical operation status of the mining electromechanical equipment.

[0013] Furthermore, the knowledge in the coal mine vertical field is acquired by using web crawler technology, data API interface or data set, including but not limited to technical documents of equipment manufacturers, operation manuals, historical fault reports, maintenance records, and information obtained through expert interviews and training materials.

[0014] Furthermore, the real-time operation data of the coal mine equipment is acquired through sensors and signal collectors, including vibration, temperature, pressure or other analog signals of the mining equipment that can characterize the operating status of the mining equipment. The sensors and signal collectors are equipped with built-in analog-to-digital converters.

[0015] Furthermore, the preprocessing of the collected historical operation data and real-time operation data of coal mine equipment includes but is not limited to removing outliers, processing missing values, standardizing and normalizing; it is specifically implemented by the following methods: the outlier processing uses a box plot, Z-score or other statistical methods to identify and decide to delete or correct the data; the missing value processing uses the mean method, median method or interpolation method to process the data; the standardized processing of the data is represented by the following formula (1); the normalized processing of the data is represented by the following formula (2):

[0016]

[0017] In formulas (1) and (2), x represents the historical operating data of mining electromechanical equipment collected; z represents the data after standardization; μ is the mean; σ is the standard deviation; x ′ represents the normalized data; min(x) represents the minimum value in the original data; max(x) represents the maximum value in the original data.

[0018] Furthermore, the construction of the domain knowledge graph is specifically achieved through the following steps:

[0019] S21. Map coal industry knowledge and equipment historical data into entities and attributes through natural language processing and information extraction technology. Each attribute includes a numerical value or descriptive text, and vectorizes the attributes. This step is represented by the following formula (3):

[0020] E=(a1,a2,…,a n ) (3)

[0021] In formula (3), E represents a coal mine equipment entity vector, a nRepresents the nth attribute of the coal mining equipment entity E;

[0022] S22. Obtain the relationship between entities through semantic analysis and relationship extraction technology, model and link the relationship, and build a knowledge graph. This step is represented by the following formula (4):

[0023] R(E1,E2)=ω (4)

[0024] In formula (4), E1 and E2 represent two entities, R represents the relationship between the entities, such as subordination or causality, and ω is a weight value, which represents the strength or importance of the relationship between the two entities.

[0025] Furthermore, in step S3, the deep learning method uses a Transformer-based large model for fine-tuning, including but not limited to:

[0026] Model fine-tuning and verification, through test scenarios simulating coal mine environments, checking the model's performance when processing actual operating data, including detailed analysis of the model's prediction results to ensure that performance fluctuations and potential failures of equipment can be accurately identified;

[0027] Performance evaluation, using standard evaluation metrics to measure the performance of the model under different operating conditions, helping to identify the strengths and weaknesses of the model and providing a basis for subsequent optimization;

[0028] Error analysis: conduct in-depth analysis of the model’s prediction errors, identify the causes of the errors, and further adjust the model parameters and training strategies to reduce the errors;

[0029] Loss function optimization: adjust the loss function of the model according to the results of error analysis to optimize the learning process;

[0030] Iterative fine-tuning: Based on the feedback from the previous steps, the model is iteratively fine-tuned.

[0031] Furthermore, when the equipment performance prediction model predicts a potential failure, the decision support logic for triggering an alarm and providing maintenance suggestions is expressed by the following formula (5):

[0032]

[0033] In formula (5), Action represents maintenance recommendations, RiskScore represents the fault risk score calculated by the equipment performance prediction model, Threshold represents the threshold of the risk score, and otherwise represents other situations.

[0034] Furthermore, the fault risk score is calculated by the following formula (6):

[0035] Risk Score=Pfault ×I impact (6)

[0036] In formula (6): P fault It is the probability of equipment failure predicted by the model. This probability value is obtained by converting the performance prediction result output by the equipment performance prediction model through the sigmoid function. impact It is the estimated severity of the impact that a failure may cause, which is usually preset by historical data and expert systems. During the specific implementation process, the estimated severity classification can be obtained based on the industry large model knowledge base.

[0037] It can be seen from the above technical solutions that this invention has the following beneficial effects:

[0038] 1. The equipment performance prediction big model proposed in the present invention integrates the historical operation data of coal mine equipment and the knowledge of the coal mine vertical field, and uses the big model technology for fine-tuning to adapt to the specific environment and needs of the mining industry. It can obtain richer industry-specific knowledge based on deep learning, improve the accuracy and sensitivity of prediction, thereby reducing false positives and false negatives, and optimizing maintenance scheduling;

[0039] 2. It can obtain the operation data of coal mine equipment in real time, and the equipment performance prediction model can immediately issue an early warning when the equipment shows the initial signs of failure, which significantly improves the response speed to failures, helps to solve problems in time, prevents the expansion of failures, and greatly reduces maintenance costs and equipment downtime;

[0040] 3. The equipment performance prediction model of the present invention can not only predict faults, but also intelligently generate maintenance suggestions based on the fault type and historical maintenance data. The automated solution push reduces the reliance on professional knowledge, allowing non-experts to quickly respond to and handle faults, thereby improving the work efficiency of the maintenance team.

[0041] Through the above-mentioned innovations, the equipment performance prediction method based on the large model of the vertical field of coal mines proposed in the present invention has significantly improved the level of intelligent management of mine equipment, and brought fundamental improvements and efficiency improvements to the operation and maintenance management of mining equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the steps of the diagnostic method of the present invention. DETAILED DESCRIPTION

[0043] A preferred embodiment of the present invention is described in detail below with reference to the accompanying drawings.

[0044] like Figure 1 The equipment performance prediction method based on the large model of the vertical field of coal mines shown includes the following steps:

[0045] S1. Collect historical operation data of coal mine equipment and pre-process the collected data.

[0046] The historical operation data of the coal mine equipment described in this preferred embodiment is obtained from the equipment monitoring system, and the historical operation data of the mining electromechanical equipment includes reliability data, maintenance records, fault types and frequencies or other data that can characterize the historical operation status of the mining electromechanical equipment.

[0047] Specifically, the preprocessing of the collected historical operation data of coal mine equipment includes but is not limited to removing outliers, processing missing values, standardizing and normalizing; it is specifically implemented by the following methods: the outlier processing uses a box plot, Z-score or other statistical methods to identify and decide to delete or correct the data; the missing value processing uses the mean method, median method or interpolation method to process the data; the standardized processing of the data is represented by the following formula (1); the normalized processing of the data is represented by the following formula (2):

[0048]

[0049] In formulas (1) and (2), x represents the historical operating data of mining electromechanical equipment collected; z represents the data after standardization; μ is the mean; σ is the standard deviation; x ′ represents the normalized data; min(x) represents the minimum value in the original data; max(x) represents the maximum value in the original data.

[0050] S2. Combine the preprocessed historical operation data of coal mine equipment with the knowledge in the vertical field of coal mines to construct a domain knowledge graph.

[0051] The coal mine vertical field knowledge is acquired by web crawler technology, data API interface or data set, including but not limited to the equipment manufacturer's technical documents, operation manuals, historical fault reports, maintenance records, and information obtained through expert interviews and training materials; the construction of the field knowledge graph is specifically achieved through the following steps:

[0052] S21. Map coal industry knowledge and equipment historical data into entities and attributes through natural language processing and information extraction technology. Each attribute includes a numerical value or descriptive text, and vectorizes the attributes. This step is represented by the following formula (3):

[0053] E=(a1,a2,…,a n ) (3)

[0054] In formula (3), E represents a coal mine equipment entity vector, a nRepresents the nth attribute of the coal mine equipment entity E; such as "equipment type", "maintenance cycle", etc.;

[0055] S22. Obtain the relationship between entities through semantic analysis and relationship extraction technology, model and link the relationship, and build a knowledge graph. This step is represented by the following formula (4):

[0056] R(E1,E2)=ω (4)

[0057] In formula (4), E1 and E2 represent two entities; R represents the relationship between entities. The relationship between entities is defined as a directed edge from one entity to another, which can be used to represent the subordination or causal relationship between entities; ω is a weight value, which represents the strength or importance of the relationship between two entities. For example, the failure mode E1 of a device may directly affect the maintenance measure E2, and the weight ω represents the urgency or degree of impact of this impact.

[0058] Taking the industry knowledge "large lump coal is coal with a particle size greater than 50 mm" as an example, in the process of constructing the knowledge graph, it is divided into the following three parts through semantic induction: "coal type: large lump coal"; "size: greater than 50 mm"; "attribute: particle size coal". This preferred embodiment extracts different types of text information to construct a knowledge graph, and obtains a basic large model in the industry field for subsequent coal mine equipment performance monitoring and early warning.

[0059] S3. Based on the general large model, the domain knowledge graph is combined to train the model, and deep learning methods are used for fine-tuning. Several labeled data are used for supervised learning to obtain a large model for equipment performance prediction and deploy it in practical applications.

[0060] The domain knowledge graph described in this preferred embodiment can help organize and understand the modeling content during the model training process. Specifically, with the help of the domain knowledge graph, the model can organize the collected data and knowledge more systematically and scientifically, so as to conduct in-depth analysis and understanding; at the same time, through the domain knowledge graph, the model can not only identify the basic status of the equipment, but also infer possible failure modes based on the operating conditions and historical failure records of the equipment; in addition, the domain knowledge graph can also provide a decision support path for the model. For example, when a specific warning signal is detected, the model can quickly find relevant treatment measures and alternative plans.

[0061] Furthermore, the label data can be obtained and used from a label database, and the deep learning method uses a large model based on Transformer for fine-tuning, specifically including:

[0062] Model fine-tuning and verification, that is, verifying the accuracy and stability of the model, by simulating test scenarios in coal mine environments to check the performance of the model when processing actual working condition data, including detailed analysis of the model's prediction results to ensure that performance fluctuations and potential failures of equipment can be accurately identified;

[0063] Performance evaluation, using standard evaluation metrics such as accuracy, recall, and F1 score to measure the performance of the model under different operating conditions, helping to identify the strengths and weaknesses of the model and provide a basis for subsequent optimization;

[0064] Error analysis: conducts in-depth analysis of the model’s prediction errors to identify the causes of the errors, including data quality issues, insufficient model structure, or inadequate feature representation. Through analysis, model parameters and training strategies can be further adjusted to reduce errors.

[0065] Loss function optimization: adjust the loss function of the model according to the results of error analysis to optimize the learning process. For example, you can adjust the weights in the loss function or add regularization terms to prevent overfitting and ensure the generalization ability of the model in practical applications.

[0066] Iterative fine-tuning: Based on the feedback from the previous steps, iteratively fine-tune the model. Specifically, according to the results of performance evaluation and error analysis, adjust the model's structure, learning rate, and other key parameters to gradually improve the model's performance in the actual environment.

[0067] S4. Collect real-time operation data of coal mine equipment and perform preprocessing including cleaning, standardization and normalization.

[0068] The real-time operation data of the coal mine equipment described in this preferred embodiment is acquired through sensors and signal collectors, including vibration, temperature, pressure or other analog signals of the mining equipment that can characterize the operating status of the mining equipment. The sensors and signal collectors are equipped with analog-to-digital converters. The analog-to-digital converter (ADC) is a device that converts continuous signals in analog form into discrete signals in digital form. Because most modern digital systems and algorithms cannot directly process the analog signals generated by sensors, these analog signals can be converted into digital signals through analog-to-digital converters, so that the sensor data can be read, processed and analyzed by computers, microprocessors or other digital devices. The preprocessing of the collected real-time operation data of the coal mine equipment is the same as the processing method described in step S1.

[0069] Specifically, the operating temperature range of the mine equipment sensor is: "-40-+60 degrees Celsius", the average relative humidity is 95%, and the output signal error is "0%-10%". In order to comprehensively predict the performance of equipment for different types of data at the same time, it is necessary to adjust the different data ranges to the range of (0-1) according to the numerical range and value of each data using standardization and normalization methods.

[0070] S5. Deploy the equipment performance prediction model into actual applications, input the pre-processed real-time operation data of coal mine equipment for performance prediction, and trigger an alarm and provide maintenance suggestions when the model predicts a potential failure.

[0071] Specifically, when the equipment performance prediction model predicts a potential failure, the decision support logic for triggering an alarm and providing maintenance suggestions is expressed by the following formula (5):

[0072]

[0073] In formula (5), Action represents maintenance recommendations; RiskScore represents the fault risk score calculated by the equipment performance prediction model; Threshold represents the threshold of the risk score; otherwise represents other situations;

[0074] The failure risk score is calculated using the following formula (6):

[0075] Risk Score=P fault ×I impact (6)

[0076] In formula (6): P fault It is the probability of equipment failure predicted by the model. This probability value is obtained by converting the performance prediction result output by the equipment performance prediction model through the sigmoid function. impact It is the estimated severity of the impact that a failure may cause, which is usually preset by historical data and expert systems. During the specific implementation process, the estimated severity classification can be obtained based on the industry large model knowledge base.

[0077] Specifically, the key monitoring parameters of the elevator include motor temperature, motor speed, voltage and load; historical data shows that when the motor temperature rises abnormally or the voltage drops suddenly, the elevator has a higher failure rate.

[0078] In the specific operation, the following features are extracted from the real-time monitoring data of the elevator:

[0079] (1) Motor temperature (TTT): An average temperature exceeding 80°C is considered abnormal;

[0080] (2) Motor speed (SSS): 10% below the standard operating speed is considered abnormal;

[0081] (3) Voltage (VVV): 5% lower than the standard voltage is abnormal;

[0082] (4) Load (LLL): Exceeding 90% of the maximum safe load is considered abnormal;

[0083] At the same time, based on historical maintenance records and the consequences of equipment failure, define the severity of the failure. impact , and set the corresponding risk score threshold; for example:

[0084] Abnormal temperature may cause motor damage, which is costly to repair. Definition I impact =0.8, Risk Score threshold is 0.5;

[0085] Voltage anomalies may cause downtime, but the cost of repair is low. Definition I impact =0.4, and the Risk Score threshold is 0.7.

[0086] In a certain monitoring, the motor temperature was abnormal. After the equipment operation data was input into the model, the model predicted the failure probability to be P fault =0.7, then the risk score is: Risk Score = 0.7 × 0.8 = 0.56.

[0087] Risk score explanation: With a risk score of 0.56 and a risk threshold of 0.5, the system will trigger an alarm and automatically push fault handling suggestions, such as reducing load or performing emergency maintenance checks.

[0088] The above-described embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for predicting equipment performance based on a large model of a coal mine vertical field, characterized in that: The following steps are involved: S1. Collect historical operation data of coal mine equipment and pre-process the collected data; S2, combining the pre-processed historical operation data of coal mine equipment with the knowledge in the vertical field of coal mines to construct a domain knowledge graph; S3. Based on the general large model, the domain knowledge graph is combined to train the model, and the deep learning method is used for fine-tuning. A number of labeled data are used for supervised learning to obtain a large model for device performance prediction and deploy it in practical applications. S4, collect real-time operation data of coal mine equipment and perform preprocessing; S5. Input the preprocessed real-time operation data of coal mine equipment into the equipment performance prediction model for performance prediction. When the model predicts a potential failure, it triggers an alarm and provides maintenance suggestions.

2. According to claim 1, a method for predicting equipment performance based on a large model of a coal mine vertical field, characterized in that: The historical operation data of the coal mine equipment is obtained from the equipment monitoring system. The historical operation data of the mining electromechanical equipment includes reliability data, maintenance records, fault types and frequencies, or other data that can characterize the historical operation status of the mining electromechanical equipment.

3. According to claim 1, a method for predicting equipment performance based on a large model of a vertical field of a coal mine, characterized in that: The knowledge in the coal mine vertical field is acquired by using web crawler technology, data API interface or data set, including but not limited to technical documents of equipment manufacturers, operation manuals, historical fault reports, maintenance records and information obtained through expert interviews and training materials.

4. According to claim 1, a method for predicting equipment performance based on a large model of a coal mine vertical field, characterized in that: The real-time operation data of the coal mine equipment is acquired through sensors and signal collectors, including vibration, temperature, pressure or other analog signals of the mining equipment that can characterize the operating status of the mining equipment. The sensors and signal collectors are equipped with analog-to-digital converters.

5. The equipment performance prediction method based on the large model of the vertical field of coal mine according to claim 1 is characterized in that: The preprocessing of the collected historical operation data and real-time operation data of coal mine equipment includes but is not limited to removing outliers, processing missing values, standardizing and normalizing. It is specifically implemented by the following methods: the outlier processing uses a box plot, Z-score or other statistical methods to identify and decide to delete or correct data; the missing value processing uses the mean method, median method or interpolation method to process the data; the standardized processing of the data is represented by the following formula (1); the normalized processing of the data is represented by the following formula (2): In formulas (1) and (2), x represents the historical operating data of mining electromechanical equipment collected; z represents the data after standardization; μ is the mean; σ is the standard deviation; x ′ represents the normalized data; min(x) represents the minimum value in the original data; max(x) represents the maximum value in the original data.

6. The equipment performance prediction method based on the vertical field large model of coal mine according to claim 1 is characterized in that: The construction of the domain knowledge graph is specifically achieved through the following steps: S21. Map coal industry knowledge and equipment historical data into entities and attributes through natural language processing and information extraction technology. Each attribute includes a numerical value or descriptive text, and vectorizes the attributes. This step is represented by the following formula (3): <h2 style=";text-align:left;direction:ltr">E = (a1,a2,…,a)<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ) (3) In formula (3), E represents a coal mining equipment entity vector, a n Represents the nth attribute of the coal mining equipment entity E; S22. Obtain the relationship between entities through semantic analysis and relationship extraction technology, model and link the relationship, and build a knowledge graph. This step is represented by the following formula (4): R(E1,E2)=ω (4) In formula (4), E1 and E2 represent two entities, R represents the relationship between the entities, such as subordination or causality, and ω is a weight value, which represents the strength or importance of the relationship between the two entities.

7. The equipment performance prediction method based on the large model of the vertical field of coal mine according to claim 1 is characterized in that: In step S3, the deep learning method uses a Transformer-based large model for fine-tuning, including but not limited to: Model fine-tuning and verification, through test scenarios simulating coal mine environments, checking the model's performance when processing actual operating data, including detailed analysis of the model's prediction results to ensure that performance fluctuations and potential failures of equipment can be accurately identified; Performance evaluation, using standard evaluation metrics to measure the performance of the model under different operating conditions, helping to identify the strengths and weaknesses of the model and providing a basis for subsequent optimization; Error analysis: conduct in-depth analysis of the model’s prediction errors, identify the causes of the errors, and further adjust the model parameters and training strategies to reduce the errors; Loss function optimization: adjust the loss function of the model according to the results of error analysis to optimize the learning process; Iterative fine-tuning: Based on the feedback from the previous steps, the model is iteratively fine-tuned.

8. The equipment performance prediction method based on the large model of the vertical field of coal mine according to claim 1 is characterized in that: When the equipment performance prediction model predicts a potential failure, the decision support logic for triggering an alarm and providing maintenance suggestions is expressed by the following formula (5): In formula (5), Action represents maintenance recommendations, RiskScore represents the fault risk score calculated by the equipment performance prediction model, Threshold represents the threshold of the risk score, and otherwise represents other situations.

9. The equipment performance prediction method based on the large model of the vertical field of coal mine according to claim 8 is characterized in that: The failure risk score is calculated using the following formula (6): Risk Score=P fault ×I impact (6) In formula (6): P fault It is the probability of equipment failure predicted by the model. This probability value is obtained by converting the performance prediction result output by the equipment performance prediction model through the sigmoid function. impact It is the estimated severity of the impact that a failure may cause, which is usually preset by historical data and expert systems. During the specific implementation process, the estimated severity classification can be obtained based on the industry large model knowledge base.

Images