An intelligent scheduling method for coal mining based on artificial intelligence platform

Abstract

The present invention discloses an intelligent scheduling method for coal mining based on an artificial intelligence platform, comprising the following steps: S1, constructing a data set; S2, preprocessing the data set; S3, analyzing the preprocessed data set to generate analysis results; S4, based on the analysis results, using variational autoencoders and graph neural networks to construct and train a prediction model; S5, using multi-agent deep reinforcement learning, combined with a hybrid expert model and a distributed deep deterministic policy gradient, to construct and train an optimization model; S6, using the prediction model to predict equipment failures, generate prediction results, give early warnings and take measures; S7, using the optimization model, dynamically adjusting the production plan according to the prediction results and real-time data, performing resource scheduling, and optimizing resource allocation; S8, establishing a feedback mechanism, collecting feedback information, and optimizing and improving resource scheduling strategies. The present invention uses big data analysis, deep learning, and a multi-agent system to achieve efficient and intelligent scheduling.

Description

Technical Field

[0001] The present invention relates to the technical field of coal mining, and in particular to an intelligent scheduling method for coal mining based on an artificial intelligence platform. Background Art

[0002] In the coal mining process, the effectiveness and safety of the dispatching system are crucial to the production efficiency of the mine and the safety of workers. Traditional coal mining dispatching methods mainly rely on manual experience and simple rules to manage resources and equipment. These methods have many shortcomings, such as untimely response to environmental changes, inability to dynamically adjust production plans, difficulty in effectively predicting equipment failures, and lack of flexible response mechanisms in emergency situations. With the development of science and technology, more and more sensor technologies and information technologies are being applied to coal mine production, but the existing dispatching systems still fail to fully utilize the advantages brought by these technologies.

[0003] In the existing technology, the Internet of Things technology has been applied to coal mine production to a certain extent for real-time monitoring of mine environment and equipment status. By installing gas sensors, temperature sensors, humidity sensors, pressure sensors and other equipment in the mine, environmental parameter data can be collected in real time. These data are transmitted to the data center via wireless networks for processing and analysis, thereby helping managers understand the real-time status of the mine. However, these systems can often only provide simple real-time monitoring and alarm functions. For complex scheduling tasks and resource optimization, they still rely on manual decision-making, making it difficult to achieve automation and intelligence.

[0004] In addition, with the development of big data technology, coal mining companies have begun to try to use big data analysis technology to analyze production data in order to improve production efficiency and safety. However, due to the complexity of the coal mine production environment and the diversity of data types, traditional big data analysis technology has many challenges in data preprocessing, feature extraction and model building.

[0005] In terms of prediction models, existing technologies usually use traditional statistical methods or simple machine learning algorithms, such as linear regression, decision trees, etc., which perform poorly when dealing with nonlinear and high-dimensional data. In recent years, deep learning technology has shown powerful capabilities in various fields, especially in dealing with complex nonlinear relationships and large-scale data. However, existing coal mine dispatching systems rarely apply advanced deep learning algorithms, such as variational autoencoders (VAE) and graph neural networks (GNN), and cannot fully utilize the powerful modeling and prediction capabilities of these algorithms.

[0006] In terms of optimization models, traditional coal mining scheduling methods usually rely on heuristic algorithms and simple optimization techniques, such as genetic algorithms and simulated annealing algorithms. These methods work well in solving simple optimization problems, but they show great limitations when faced with complex multi-objective optimization problems and dynamically changing production environments. In recent years, the development of multi-agent systems and deep reinforcement learning technologies has provided new ideas for solving complex optimization problems. However, existing coal mine scheduling systems rarely introduce multi-agent systems and advanced optimization algorithms, and cannot achieve global optimization and dynamic scheduling of resources.

[0007] Therefore, how to provide an intelligent scheduling method for coal mining based on an artificial intelligence platform is an urgent problem that technicians in this field need to solve. Summary of the invention

[0008] One purpose of the present invention is to propose an intelligent scheduling method for coal mining based on an artificial intelligence platform. The present invention makes full use of the Internet of Things, big data analysis, variational autoencoders, graph neural networks and multi-agent system technologies to construct an efficient intelligent scheduling method for coal mining. The method realizes dynamic allocation of resources and early warning of equipment failures through real-time data collection and processing, accurate prediction and optimized scheduling. It has the advantages of high efficiency, safety, adaptability and flexibility, greatly improves the production efficiency and safety of coal mines, and provides a strong guarantee for the efficient operation of mines.

[0009] According to an embodiment of the present invention, a coal mining intelligent scheduling method based on an artificial intelligence platform includes the following steps:

[0010] S1. Use IoT technology to connect and collect sensor data in mines and build data sets;

[0011] S2, preprocessing the data set, including data cleaning and removing noise and outliers;

[0012] S3. Analyze the preprocessed data set using big data analysis technology to generate analysis results;

[0013] S4. Based on the analysis results, a prediction model for coal mining is constructed and trained using variational autoencoders and graph neural networks.

[0014] S5. Use a multi-agent system, combined with a hybrid expert model and distributed deep deterministic policy gradient to build and train an optimization model for coal mining;

[0015] S6. Use the prediction model to predict equipment failures, generate prediction results, issue early warnings, and take measures;

[0016] S7. Use the optimization model to dynamically adjust the production plan, perform resource scheduling, and optimize resource allocation based on the forecast results and real-time data;

[0017] S8. Establish a feedback mechanism, collect feedback information, and optimize and improve resource scheduling strategies.

[0018] Furthermore, the S3 specifically includes:

[0019] S31. Format the preprocessed data set and convert the data into a structured format to ensure that the data can be processed by big data analysis tools;

[0020] S32, extracting key features from the formatted data set;

[0021] S33. Use principal component analysis to select key features that affect coal mining and generate feature vectors ,in represents the i-th feature;

[0022] S34, aggregate the feature data in the same time window, generate a time series data set, and set the time window Generate time series for aggregation period ,in represents the jth time point;

[0023] S35, use the RobustScaler-based method to normalize the eigenvector X, and convert the eigenvalue Convert to normalized value :

[0024] ;

[0025] in, Representation characteristics The 25th percentile of Representation characteristics The 75th percentile of the method is more robust when dealing with data with outliers.

[0026] S36. Use Apache Spark to perform preliminary analysis on the normalized data and generate statistical reports and data visualization charts;

[0027] S37. Based on the preliminary analysis results, prepare the data set for modeling and generate the training set and test set.

[0028] Furthermore, the S4 specifically includes:

[0029] S41, constructing an initial feature vector using the training set of S37;

[0030] S42, using a variational autoencoder to perform dimensionality reduction processing on the initial feature matrix to generate a low-dimensional latent variable z;

[0031] S43. Define the loss function of the variational autoencoder as the sum of the reconstruction error and the KL divergence:

[0032] ;

[0033] in, represents the loss function of the variational autoencoder, represents the conditional probability distribution, represents the approximate posterior distribution, represents the prior distribution, E represents the expected value, represents the KL divergence, represents the parameters of the encoder, Represents the parameters of the decoder;

[0034] S44. Based on the latent variable z after dimensionality reduction, the graph neural network is used for relationship modeling to construct the graph structure G=(V,E) of mine mining, where the node set V represents different equipment and resources, and the edge set E represents the node relationship;

[0035] S45. In the graph neural network, a graph convolutional network is used to extract features and generate embedding vectors for the graph structure G. The inter-layer propagation rule of the graph convolutional network is defined as:

[0036] ;

[0037] in, Indicates The node feature matrix of the layer, Indicates The node feature matrix of the layer, A represents the adjacency matrix, D is the degree matrix of A, Indicates The weight matrix of the layer, represents a nonlinear activation function;

[0038] S46. Use the graph attention network to process the node embedding vector H generated by the graph convolutional network, and calculate the attention coefficient and attention weight update rule between nodes:

[0039] ;

[0040] in, represents the attention energy between node i and node j, LeakyReLU represents the activation function, represents the parameter vector of the attention mechanism, W represents the weight matrix, represents the embedding vector of node i, represents the embedding vector of node j, Represents a vector concatenation operation;

[0041] ;

[0042] in, represents the attention coefficient between node i and node j, and N(i) represents the set of neighbor nodes of node i;

[0043] S47. In the graph attention network, a multi-head attention mechanism is introduced to improve the stability and expressiveness of the model. The multi-head attention mechanism is defined as:

[0044] ;

[0045] in, represents the vector concatenation operation, K represents the number of attention heads, represents the attention coefficient of the kth attention head, represents the weight matrix of the kth attention head, Represents the updated node embedding vector;

[0046] S48, embed the hidden variable z and the node generated by the graph convolutional network into a vector Combined with the above, a prediction model for coal mining is constructed. The prediction model receives the input feature vector and outputs the prediction result. ;

[0047] S49. Define the loss function of the prediction model as a multi-task loss function:

[0048] ;

[0049] in, represents the loss function of the prediction model, N represents the number of samples, Indicates the actual value, represents the predicted value, represents the regularization coefficient, W represents the model parameter matrix, represents the weight of contrastive learning loss, P represents the set of positive sample pairs, and k represents the negative sample;

[0050] S410, minimize the prediction model through back propagation algorithm and gradient descent optimization algorithm The loss function is trained to obtain the final prediction model for coal mining.

[0051] Furthermore, the S5 specifically includes:

[0052] S51. Deploy multiple agents in the mine, each agent being responsible for different resource and equipment scheduling tasks, wherein the resources and equipment include mining equipment, transportation equipment, ventilation equipment, drainage equipment, power equipment, safety equipment, personnel scheduling and auxiliary equipment;

[0053] S52. The intelligent agent perceives the status of resources and environmental information through the Internet of Things technology, including collecting the operating status of excavators and drills, the transportation status of mine cars and belt conveyors, the ventilation status of fans and dampers, the drainage status of water pumps, the power status of transformers and cables, the safety status of gas detectors and roof monitoring systems, the operating status of miners and safety officers, and the usage status of maintenance tools and material supplies;

[0054] S53, each agent makes a decision based on a hybrid expert model, wherein the hybrid expert model includes multiple expert sub-models. The agent selects the optimal expert sub-model for scheduling decision according to the current environment state and task requirements:

[0055] ;

[0056] Among them, y represents the final output, N represents the number of expert sub-models, represents the weight of the gating network output, represents the output of the i-th expert sub-model, and x represents the input feature;

[0057] Defining the weights of the gating network :

[0058] ;

[0059] in, represents the weight matrix of the i-th gating network, Represents the bias value of the i-th gating network;

[0060] S54, the agent optimizes the scheduling strategy through a distributed deep deterministic policy gradient, wherein the distributed deep deterministic policy gradient includes a policy network and value network , where s represents the state and a represents the action. represents the parameters of the policy network, Parameters representing the value network;

[0061] S55. The agent optimizes the parameters of the strategy network and value network by continuously interacting with the environment:

[0062] ;

[0063] ;

[0064] in, represents the learning rate of the policy network, represents the learning rate of the value network, r represents the reward, represents the discount factor, Representation strategy The state distribution under , s' represents the next state, Parameters representing the target value grid;

[0065] S56, the intelligent agents share the environmental status and scheduling strategy through the communication network to form a global optimization plan, which specifically includes the intelligent agents exchanging the environmental status information they perceive and the current scheduling decision to improve the resource utilization efficiency and scheduling optimization level of the overall system;

[0066] S57. Based on the shared global environmental status and scheduling strategy, the intelligent agent dynamically adjusts the production plan and resource scheduling, including optimizing the scheduling of mining equipment, transportation equipment, ventilation equipment, drainage equipment, power equipment and safety equipment in the mine according to real-time environmental data and prediction results, so as to improve production efficiency and system flexibility.

[0067] Furthermore, the S8 specifically includes:

[0068] S81. Establish a data feedback system, install multiple data collection terminals at the mine site, monitor the environmental parameters, equipment operation status and production progress in the mine in real time, and transmit the collected data to the central data processing platform through the Internet of Things technology;

[0069] S82. Classify and store the collected data using a database management system, including structured data and unstructured data, using a MySQL database to store structured data and a MongoDB database to store unstructured data;

[0070] S83. Use big data analysis tools to perform batch and stream processing on stored data, extract key features and trend information, and mark and classify abnormal data;

[0071] S84. Update and retrain the existing prediction model and optimization model based on the feedback data:

[0072] ;

[0073] in, represents the loss function, N represents the number of samples, Indicates the actual value, represents the model prediction value, represents the model parameters;

[0074] S85. According to the updated model output results, the mine production plan and resource scheduling strategy are adjusted, and a genetic algorithm is used for multi-objective optimization;

[0075] S86. Display the processed data and adjusted scheduling strategies to mine managers through visualization tools, providing a real-time monitoring interface and decision support system.

[0076] The beneficial effects of the present invention are:

[0077] First, the IoT technology is used to comprehensively collect sensor data in the mine, and a data set with wide coverage and strong real-time performance is constructed. By preprocessing the data, the accuracy and completeness of the data are ensured, solving the problem of imperfect data collection and processing in the existing technology. This provides a solid foundation for subsequent data analysis and model construction, enabling the system to accurately reflect the real-time status of the mine.

[0078] Secondly, big data analysis technology is used to conduct in-depth analysis of the preprocessed data to extract key features and trend information. By using advanced deep learning algorithms such as variational autoencoders and graph neural networks, efficient prediction models are built and trained. These algorithms can handle complex nonlinear relationships and large-scale data, improve the prediction accuracy and robustness of the model, and enable the system to more accurately predict changes in the mine environment and equipment status, timely detect potential problems and provide early warnings.

[0079] In terms of optimization scheduling, the present invention introduces a multi-agent system, combines a hybrid expert model and a distributed deep deterministic policy gradient, and constructs and trains an optimization model. Multiple agents are responsible for different resource and equipment scheduling tasks, and through collaboration and information sharing, global optimization and dynamic scheduling of resources are achieved. This method not only improves the flexibility and response speed of production planning, but also optimizes resource allocation and maximizes production efficiency. At the same time, the hybrid expert model can select the optimal expert sub-model for decision-making based on the current environmental status and task requirements, ensuring the scientificity and effectiveness of the scheduling strategy.

[0080] In terms of equipment failure prediction, the prediction model is used to monitor and analyze the operating status of the equipment in real time, generate prediction results and provide early warning. This method can effectively reduce unplanned downtime, improve equipment utilization and production continuity, significantly reduce maintenance costs and production risks, and ensure the safe and stable operation of coal mine production.

[0081] In addition, by establishing a feedback mechanism, collecting feedback information from the production process in real time, and using big data analysis and adaptive learning algorithms to update and retrain the prediction model and optimization model, the system can continuously optimize and improve the scheduling strategy. This closed-loop feedback mechanism ensures the self-learning and adaptive capabilities of the system, allowing the scheduling system to continuously adapt to changes in the mine environment and production conditions, and continuously improve the level of scheduling optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0083] Figure 1 This is a flow chart of an intelligent scheduling method for coal mining based on an artificial intelligence platform proposed by the present invention;

[0084] Figure 2 A schematic diagram of a multi-agent system of an intelligent scheduling method for coal mining based on an artificial intelligence platform proposed by the present invention;

[0085] Figure 3 This is a feedback mechanism and model update flow chart of an intelligent scheduling method for coal mining based on an artificial intelligence platform proposed by the present invention. DETAILED DESCRIPTION

[0086] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0087] refer to Figure 1-3 , an intelligent scheduling method for coal mining based on an artificial intelligence platform, comprising the following steps:

[0088] S1. Use IoT technology to connect and collect sensor data in mines and build data sets;

[0089] S2, preprocessing the data set, including data cleaning and removing noise and outliers;

[0090] S3. Analyze the preprocessed data set using big data analysis technology to generate analysis results;

[0091] S4. Based on the analysis results, a prediction model for coal mining is constructed and trained using variational autoencoders and graph neural networks.

[0092] S5. Use a multi-agent system, combined with a hybrid expert model and distributed deep deterministic policy gradient to build and train an optimization model for coal mining;

[0093] S6. Use the prediction model to predict equipment failures, generate prediction results, issue early warnings, and take measures;

[0094] S7. Use the optimization model to dynamically adjust the production plan, perform resource scheduling, and optimize resource allocation based on the forecast results and real-time data;

[0095] S8. Establish a feedback mechanism, collect feedback information, and optimize and improve resource scheduling strategies.

[0096] In this implementation, S3 specifically includes:

[0097] S31. Format the preprocessed data set and convert the data into a structured format to ensure that the data can be processed by big data analysis tools;

[0098] S32, extracting key features from the formatted data set;

[0099] S33. Use principal component analysis to select key features that affect coal mining and generate feature vectors ,in represents the i-th feature;

[0100] S34, aggregate the feature data in the same time window, generate a time series data set, and set the time window Generate time series for aggregation period ,in represents the jth time point;

[0101] S35, use the RobustScaler-based method to normalize the eigenvector X, and convert the eigenvalue Convert to normalized value :

[0102] ;

[0103] in, Representation characteristics The 25th percentile of Representation characteristics The 75th percentile of the method is more robust when dealing with data with outliers;

[0104] S36. Use Apache Spark to perform preliminary analysis on the normalized data and generate statistical reports and data visualization charts;

[0105] S37. Based on the preliminary analysis results, prepare the data set for modeling and generate the training set and test set.

[0106] In this implementation manner, the S4 specifically includes:

[0107] S41, constructing an initial feature vector using the training set of S37;

[0108] S42, using a variational autoencoder to perform dimensionality reduction processing on the initial feature matrix to generate a low-dimensional latent variable z;

[0109] S43. Define the loss function of the variational autoencoder as the sum of the reconstruction error and the KL divergence:

[0110] ;

[0111] in, represents the loss function of the variational autoencoder, represents the conditional probability distribution, represents the approximate posterior distribution, represents the prior distribution, E represents the expected value, represents the KL divergence, represents the parameters of the encoder, Represents the parameters of the decoder;

[0112] S44. Based on the latent variable z after dimensionality reduction, the graph neural network is used for relationship modeling to construct the graph structure G=(V,E) of mine mining, where the node set V represents different equipment and resources, and the edge set E represents the node relationship;

[0113] S45. In the graph neural network, a graph convolutional network is used to extract features and generate embedding vectors for the graph structure G. The inter-layer propagation rule of the graph convolutional network is defined as:

[0114] ;

[0115] in, Indicates The node feature matrix of the layer, Indicates The node feature matrix of the layer, A represents the adjacency matrix, D is the degree matrix of A, Indicates The weight matrix of the layer, represents a nonlinear activation function;

[0116] S46. Use the graph attention network to process the node embedding vector H generated by the graph convolutional network, and calculate the attention coefficient and attention weight update rule between nodes:

[0117] ;

[0118] in, represents the attention energy between node i and node j, LeakyReLU represents the activation function, represents the parameter vector of the attention mechanism, W represents the weight matrix, represents the embedding vector of node i, represents the embedding vector of node j, Represents a vector concatenation operation;

[0119] ;

[0120] in, represents the attention coefficient between node i and node j, and N(i) represents the set of neighbor nodes of node i;

[0121] S47. In the graph attention network, a multi-head attention mechanism is introduced to improve the stability and expressiveness of the model. The multi-head attention mechanism is defined as:

[0122] ;

[0123] in, represents the vector concatenation operation, K represents the number of attention heads, represents the attention coefficient of the kth attention head, represents the weight matrix of the kth attention head, Represents the updated node embedding vector;

[0124] S48, embed the hidden variable z and the node generated by the graph convolutional network into a vector Combined with the above, a prediction model for coal mining is constructed. The prediction model receives the input feature vector and outputs the prediction result. ;

[0125] S49. Define the loss function of the prediction model as a multi-task loss function:

[0126] ;

[0127] in, represents the loss function of the prediction model, N represents the number of samples, Indicates the actual value, represents the predicted value, represents the regularization coefficient, W represents the model parameter matrix, represents the weight of contrastive learning loss, P represents the set of positive sample pairs, and k represents the negative sample;

[0128] S410, minimize the prediction model through back propagation algorithm and gradient descent optimization algorithm The loss function is trained to obtain the final prediction model for coal mining.

[0129] In this implementation manner, S5 specifically includes:

[0130] S51. Deploy multiple agents in the mine, each agent being responsible for different resource and equipment scheduling tasks, wherein the resources and equipment include mining equipment, transportation equipment, ventilation equipment, drainage equipment, power equipment, safety equipment, personnel scheduling and auxiliary equipment;

[0131] S52. The intelligent agent perceives the status of resources and environmental information through the Internet of Things technology, including collecting the operating status of excavators and drills, the transportation status of mine cars and belt conveyors, the ventilation status of fans and dampers, the drainage status of water pumps, the power status of transformers and cables, the safety status of gas detectors and roof monitoring systems, the operating status of miners and safety officers, and the usage status of maintenance tools and material supplies;

[0132] S53, each agent makes a decision based on a hybrid expert model, wherein the hybrid expert model includes multiple expert sub-models. The agent selects the optimal expert sub-model for scheduling decision according to the current environment state and task requirements:

[0133] ;

[0134] Among them, y represents the final output, N represents the number of expert sub-models, represents the weight of the gating network output, represents the output of the i-th expert sub-model, and x represents the input feature;

[0135] Defining the weights of the gating network :

[0136] ;

[0137] in, represents the weight matrix of the i-th gating network, Represents the bias value of the i-th gating network;

[0138] S54, the agent optimizes the scheduling strategy through a distributed deep deterministic policy gradient, wherein the distributed deep deterministic policy gradient includes a policy network and value network , where s represents the state and a represents the action. represents the parameters of the policy network, Parameters representing the value network;

[0139] S55. The agent optimizes the parameters of the strategy network and value network by continuously interacting with the environment:

[0140] ;

[0141] ;

[0142] in, represents the learning rate of the policy network, represents the learning rate of the value network, r represents the reward, represents the discount factor, Representation strategy The state distribution under , s' represents the next state, Parameters representing the target value grid;

[0143] S56, the intelligent agents share the environmental status and scheduling strategy through the communication network to form a global optimization plan, which specifically includes the intelligent agents exchanging the environmental status information they perceive and the current scheduling decision to improve the resource utilization efficiency and scheduling optimization level of the overall system;

[0144] S57. Based on the shared global environmental status and scheduling strategy, the intelligent agent dynamically adjusts the production plan and resource scheduling, including optimizing the scheduling of mining equipment, transportation equipment, ventilation equipment, drainage equipment, power equipment and safety equipment in the mine according to real-time environmental data and prediction results, so as to improve production efficiency and system flexibility.

[0145] In this implementation manner, S8 specifically includes:

[0146] S81. Establish a data feedback system, install multiple data collection terminals at the mine site, monitor the environmental parameters, equipment operation status and production progress in the mine in real time, and transmit the collected data to the central data processing platform through the Internet of Things technology;

[0147] S82. Classify and store the collected data using a database management system, including structured data and unstructured data, using a MySQL database to store structured data and a MongoDB database to store unstructured data;

[0148] S83. Use big data analysis tools to perform batch and stream processing on stored data, extract key features and trend information, and mark and classify abnormal data;

[0149] S84. Update and retrain the existing prediction model and optimization model based on the feedback data:

[0150] ;

[0151] in, represents the loss function, N represents the number of samples, Indicates the actual value, represents the model prediction value, represents the model parameters;

[0152] S85. According to the updated model output results, the mine production plan and resource scheduling strategy are adjusted, and a genetic algorithm is used for multi-objective optimization;

[0153] S86. Display the processed data and adjusted scheduling strategies to mine managers through visualization tools, providing a real-time monitoring interface and decision support system.

[0154] Embodiment 1:

[0155] In order to verify the feasibility of the present invention in implementation, the present invention is applied to a large coal mine. The annual output of the coal mine is 5 million tons, the depth of the mine is 800 meters, the working environment is complex, and there are many uncertain factors in the production process, such as gas concentration, equipment failure, ventilation system failure, etc. The traditional coal mine dispatching system mainly relies on manual experience and simple rules for management, and has problems such as slow response speed, low prediction accuracy, and poor dispatching efficiency, which seriously affects production efficiency and safety.

[0156] In order to improve the production efficiency and safety of coal mines, an intelligent scheduling method for coal mining based on an artificial intelligence platform was adopted. First, a variety of sensors, including gas detection sensors, temperature sensors, humidity sensors, and pressure sensors, were installed in the mine to build a comprehensive real-time data acquisition system. These sensors are connected through the Internet of Things technology and transmit data to the central data processing platform in real time, forming a comprehensive data set.

[0157] After data collection, the system preprocessed the data set, including data cleaning, noise and outlier removal, to ensure the accuracy and completeness of the data. Then, the preprocessed data set was analyzed using big data analysis technology to extract key features and trend information and generate a detailed analysis report.

[0158] Based on the analysis results, the system uses variational autoencoders and graph neural networks to build and train prediction models for coal mining. These models can handle complex nonlinear relationships and predict changes in mine environment and equipment status. For example, the system can predict the trend of gas concentration changes in the next 24 hours, and warn of potential risks of gas exceeding the standard, so that timely measures can be taken.

[0159] In terms of optimization and scheduling, the system introduces a multi-agent system, combines a hybrid expert model and a distributed deep deterministic policy gradient, and builds and trains an optimization model. Multiple agents are responsible for different resource and equipment scheduling tasks, such as mining equipment, transportation equipment, ventilation equipment, drainage equipment, etc. Through collaboration and information sharing between agents, the system achieves global optimization and dynamic scheduling of resources. For example, when an excavator fails, the system can automatically adjust the production plan and reallocate other equipment and resources to ensure the continuity and efficiency of production.

[0160] In terms of equipment failure prediction, the system uses prediction models to monitor and analyze the operating status of equipment in real time. For example, by analyzing the vibration data and operating temperature data of a certain excavator, the system predicts that the equipment may fail in the next two days. Based on the prediction results, the system arranges maintenance in advance, avoiding unplanned downtime caused by equipment failure and reducing production losses.

[0161] ;

[0162] Analyzing the data in Table 1, it can be concluded that after applying the intelligent scheduling method for coal mining based on the artificial intelligence platform, the production efficiency and safety of coal mines have been significantly improved. First, the annual output of coal mines increased from 5 million tons in 2022 to 5.5 million tons in 2023, an increase of 10%. In terms of monthly output, it reached 450,000 tons in June 2023, while it was only 400,000 tons in the same period of 2022, with a growth rate of 12.5%. These data show that the overall production capacity of coal mines has been greatly improved by optimizing production plans and resource allocation through intelligent scheduling methods.

[0163] In terms of equipment utilization, the average utilization of excavators increased from 85% in 2022 to 98% in 2023, an increase of 15%. This significant improvement is mainly due to the precise prediction and timely maintenance arrangement of the intelligent scheduling system, which reduced the unplanned downtime of equipment. The number of equipment failures also decreased significantly, from 35 in 2022 to 10 in 2023, a decrease of 71.4%. These data prove the effectiveness of the intelligent scheduling system in equipment management and maintenance.

[0164] In terms of safety, by introducing the intelligent dispatching system, the coal mine successfully warned and avoided 12 gas exceeding the standard in 2023, while it failed to successfully warn any gas exceeding the standard in 2022. This improvement greatly improved the safety of the mine. No safety accidents occurred in 2023, while 3 safety accidents occurred in 2022, and the number of accidents was reduced by 100%. This shows the excellent performance of the intelligent dispatching system in real-time monitoring and safety warning.

[0165] In addition, the intelligent scheduling system has significantly reduced unplanned downtime, from 1,200 hours in 2022 to 600 hours in 2023, a reduction of 50%. The average downtime per equipment failure has also been reduced from 35 hours to 10 hours, a reduction of 71.4%. These data show that through intelligent scheduling and optimization, coal mines can manage production processes more efficiently, reduce downtime losses, and improve equipment utilization and production continuity.

[0166] Feedback mechanism and model update are also important components of the intelligent dispatching system. In 2023, the system conducted 15 model updates based on actual operation data, while there were no model updates in 2022. Through continuous model optimization and adjustment, the system can adapt to the changing production environment and provide more accurate and efficient dispatching strategies. In addition, after the dispatch optimization, the average monthly electricity saving reached 100,000 kWh, and the overall cost saving reached 3 million yuan, showing significant economic benefits.

[0167] In summary, the intelligent scheduling method for coal mining based on the artificial intelligence platform has significantly improved production efficiency and equipment utilization, enhanced safety, reduced unplanned downtime, and brought significant economic benefits through advanced technical means. These data fully demonstrate the great potential and value of this method in practical applications, and provide solid technical support for the intelligent and modern management of coal mines.

[0168] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent scheduling method for coal mining based on an artificial intelligence platform, characterized in that: The steps include: S1. Use IoT technology to connect and collect sensor data in mines and build data sets; S2, preprocessing the data set, including data cleaning and removing noise and outliers; S3. Analyze the preprocessed data set using big data analysis technology to generate analysis results; S4. Based on the analysis results, a prediction model for coal mining is constructed and trained using variational autoencoders and graph neural networks. S5. Use a multi-agent system, combined with a hybrid expert model and distributed deep deterministic policy gradient to build and train an optimization model for coal mining; S6. Use the prediction model to predict equipment failures, generate prediction results, issue early warnings, and take measures; S7. Use the optimization model to dynamically adjust the production plan, perform resource scheduling, and optimize resource allocation based on the forecast results and real-time data; S8. Establish a feedback mechanism, collect feedback information, and optimize and improve resource scheduling strategies; The S5 specifically includes: S51. Deploy multiple agents in the mine, each agent being responsible for different resource and equipment scheduling tasks, wherein the resources and equipment include mining equipment, transportation equipment, ventilation equipment, drainage equipment, power equipment, safety equipment, personnel scheduling and auxiliary equipment; S52. The intelligent agent perceives the status of resources and environmental information through the Internet of Things technology, including collecting the operating status of excavators and drills, the transportation status of mine cars and belt conveyors, the ventilation status of fans and dampers, the drainage status of water pumps, the power status of transformers and cables, the safety status of gas detectors and roof monitoring systems, the operating status of miners and safety officers, and the usage status of maintenance tools and material supplies; S53, each agent makes a decision based on a hybrid expert model, wherein the hybrid expert model includes multiple expert sub-models. The agent selects the optimal expert sub-model for scheduling decision according to the current environment state and task requirements: ； Among them, y represents the final output, N represents the number of expert sub-models, represents the weight of the gating network output, represents the output of the i-th expert sub-model, and x represents the input feature; Defining the weights of the gating network : ; in, represents the weight matrix of the i-th gating network, Represents the bias value of the i-th gating network; S54, the agent optimizes the scheduling strategy through a distributed deep deterministic policy gradient, wherein the distributed deep deterministic policy gradient includes a policy network and value network , where s represents the state and a represents the action. represents the parameters of the policy network, Parameters representing the value network; S55. The agent optimizes the parameters of the strategy network and value network by continuously interacting with the environment: ; ; in, represents the learning rate of the policy network, represents the learning rate of the value network, r represents the reward, represents the discount factor, Representation strategy The state distribution under , s' represents the next state, Parameters representing the target value grid; S56, the intelligent agents share the environmental status and scheduling strategy through the communication network to form a global optimization solution, which specifically includes the intelligent agents exchanging the environmental status information they perceive and the current scheduling decision; S57. Based on the shared global environmental status and scheduling strategy, the intelligent agent dynamically adjusts the production plan and resource scheduling, including optimizing the scheduling of mining equipment, transportation equipment, ventilation equipment, drainage equipment, power equipment and safety equipment in the mine according to real-time environmental data and prediction results.

2. The method for intelligent scheduling of coal mining based on an artificial intelligence platform according to claim 1, characterized in that: The S3 specifically includes: S31, formatting the preprocessed data set and converting the data into a structured format; S32, extracting key features from the formatted data set; S33. Use principal component analysis to select key features that affect coal mining and generate feature vectors ,in represents the i-th feature; S34, aggregate the feature data in the same time window, generate a time series data set, and set the time window Generate time series for the aggregation period ,in represents the jth time point; S35, use the RobustScaler-based method to normalize the eigenvector X, and convert the eigenvalue Convert to normalized value : ; in, Representation characteristics The 25th percentile of Representation characteristics 75th percentile; S36. Use Apache Spark to perform preliminary analysis on the normalized data and generate statistical reports and data visualization charts; S37. Based on the preliminary analysis results, prepare the data set for modeling and generate the training set and test set.

3. The method for intelligent scheduling of coal mining based on an artificial intelligence platform according to claim 1 is characterized in that: The S4 specifically includes: S41, constructing an initial feature vector using the training set of S37; S42, using a variational autoencoder to perform dimensionality reduction processing on the initial feature matrix to generate a low-dimensional latent variable z; S43. Define the loss function of the variational autoencoder as the sum of the reconstruction error and the KL divergence: ; in, represents the loss function of the variational autoencoder, represents the conditional probability distribution, represents the approximate posterior distribution, represents the prior distribution, E represents the expected value, represents the KL divergence, represents the parameters of the encoder, Represents the parameters of the decoder; S44. Based on the latent variable z after dimensionality reduction, the graph neural network is used for relationship modeling to construct the graph structure G=(V,E) of mine mining, where the node set V represents different equipment and resources, and the edge set E represents the node relationship; S45. In the graph neural network, a graph convolutional network is used to extract features and generate embedding vectors for the graph structure G. The inter-layer propagation rule of the graph convolutional network is defined as: ; in, Indicates The node feature matrix of the layer, Indicates The node feature matrix of the layer, A represents the adjacency matrix, D is the degree matrix of A, Indicates The weight matrix of the layer, represents a nonlinear activation function; S46. Use the graph attention network to process the node embedding vector H generated by the graph convolutional network, and calculate the attention coefficient and attention weight update rule between nodes: ; in, represents the attention energy between node i and node j, LeakyReLU represents the activation function, represents the parameter vector of the attention mechanism, W represents the weight matrix, represents the embedding vector of node i, represents the embedding vector of node j, Represents a vector concatenation operation; ; in, represents the attention coefficient between node i and node j, and N(i) represents the set of neighbor nodes of node i; S47. In the graph attention network, a multi-head attention mechanism is introduced and defined as: ; in, represents the vector concatenation operation, K represents the number of attention heads, represents the attention coefficient of the kth attention head, represents the weight matrix of the kth attention head, represents the updated node embedding vector; S48, embed the hidden variable z and the node generated by the graph convolutional network into a vector Combined with the above, a prediction model for coal mining is constructed. The prediction model receives the input feature vector and outputs the prediction result. ; S49. Define the loss function of the prediction model as a multi-task loss function: ; in, represents the loss function of the prediction model, N represents the number of samples, Indicates the actual value, represents the predicted value, represents the regularization coefficient, W represents the model parameter matrix, represents the weight of contrastive learning loss, P represents the set of positive sample pairs, and k represents the negative sample; S410, minimize the prediction model through back propagation algorithm and gradient descent optimization algorithm The loss function is trained to obtain the final prediction model for coal mining.

4. The method for intelligent scheduling of coal mining based on an artificial intelligence platform according to claim 1, characterized in that: The S8 specifically includes: S81. Establish a data feedback system, install multiple data collection terminals at the mine site, monitor the environmental parameters, equipment operation status and production progress in the mine in real time, and transmit the collected data to the central data processing platform through the Internet of Things technology; S82. Classify and store the collected data using a database management system, including structured data and unstructured data, using a MySQL database to store structured data and a MongoDB database to store unstructured data; S83. Use big data analysis tools to perform batch and stream processing on stored data, extract key features and trend information, and mark and classify abnormal data; S84. Update and retrain the existing prediction model and optimization model based on the feedback data: ; in, represents the loss function, N represents the number of samples, Indicates the actual value, represents the model prediction value, represents the model parameters; S85. According to the updated model output results, the mine production plan and resource scheduling strategy are adjusted, and a genetic algorithm is used for multi-objective optimization; S86. Display the processed data and adjusted scheduling strategies to mine managers through visualization tools, providing a real-time monitoring interface and decision support system.

Images