Prediction method and device for coal mine operation scheduling data and storage medium

Abstract

The invention provides a coal mine operation scheduling data prediction method and device and a storage medium. According to the method, original multi-source heterogeneous data of a coal mine are obtained, the original multi-source heterogeneous data are preprocessed to obtain multi-source heterogeneous data, and the original multi-source heterogeneous data comprise production data, equipment data, cost data and safety data; the multi-source heterogeneous data is input into a pre-constructed fusion model for prediction, a prediction result of the coal mine operation scheduling data is obtained, and the pre-constructed fusion model is constructed based on a large model and multiple machine models; the prediction result comprises a production yield prediction result, an equipment operation state prediction result, a cost consumption prediction result and a safety risk prediction result; and performing coal mine operation scheduling based on the prediction result. According to the invention, the problem of low prediction precision of coal mine operation scheduling data in the prior art is solved.

Description

Technical Field

[0001] This application relates to the technical field of digital management of coal mining enterprises and artificial intelligence. Specifically, it relates to a method for predicting coal mine operation and scheduling data, a device for predicting coal mine operation and scheduling data, and a computer-readable storage medium. Background Technology

[0002] Currently, coal mine enterprises mainly rely on traditional statistical models for operational scheduling data prediction. These statistical models are difficult to adapt to the complex characteristics of coal mine production, and the core problems are concentrated in the following three points:

[0003] Weak nonlinear data processing capability: Coal mine production is affected by multiple factors such as geological conditions, underground environment, and equipment aging, resulting in data with strong nonlinear characteristics. Traditional models cannot capture deep correlations, leading to large prediction errors.

[0004] Multi-source data fusion gap: Scheduling data is scattered across independent platforms such as production, equipment, cost, and security. Traditional methods can only analyze data from a single dimension, resulting in data silos.

[0005] Poor dynamic adaptability: Traditional models have fixed parameters after training. When faced with dynamic changes such as sudden equipment failures, coal price fluctuations, and policy adjustments, manual retraining is required, which takes a long time and cannot support "minute-level" scheduling decisions.

[0006] The above problems ultimately lead to low prediction accuracy of coal mine operation and scheduling data. Summary of the Invention

[0007] The main objective of this application is to provide a method, apparatus and computer-readable storage medium for predicting coal mine operation and scheduling data, so as to at least solve the problem of low prediction accuracy of coal mine operation and scheduling data in the prior art.

[0008] To achieve the above objectives, according to one aspect of this application, a method for predicting coal mine operation and scheduling data is provided, comprising: acquiring raw multi-source heterogeneous data of a coal mine, and preprocessing the raw multi-source heterogeneous data to obtain multi-source heterogeneous data, wherein the raw multi-source heterogeneous data includes production data, equipment data, cost data, and safety data; inputting the multi-source heterogeneous data into a pre-constructed fusion model for prediction to obtain prediction results of coal mine operation and scheduling data, wherein the pre-constructed fusion model is constructed based on a large model and multiple machine models, and the prediction results include production output prediction results, equipment operating status prediction results, cost consumption prediction results, and safety risk prediction results; and performing coal mine operation and scheduling based on the prediction results.

[0009] Optionally, inputting the multi-source heterogeneous data into a pre-built fusion model for prediction to obtain prediction results for coal mine operation and scheduling data includes: inputting the multi-source heterogeneous data into a first large model for prediction to obtain a first prediction result; simultaneously using the first large model to extract features from the multi-source heterogeneous data to obtain a multi-dimensional feature vector, wherein the multi-dimensional feature vector includes time-dependent features and cross-domain correlation features; inputting the multi-dimensional feature vector into a first machine learning model and a second machine learning model for prediction to obtain a second prediction result and a third prediction result, respectively; and determining the prediction result based on the first prediction result, the second prediction result, and the third prediction result.

[0010] Optionally, determining the prediction result based on the first prediction result, the second prediction result, and the third prediction result includes: performing a weighted fusion of the first prediction result, the second prediction result, and the third prediction result to obtain the prediction result, wherein the weights of the first prediction result, the second prediction result, and the third prediction result are adjusted based on the prediction error of the large model.

[0011] Optionally, before inputting the multi-source heterogeneous data into a pre-built fusion model for prediction to obtain the prediction results of coal mine operation and scheduling data, the method further includes: acquiring historical data of the coal mine, the historical data including historical generation data, historical equipment data, historical cost data, historical safety data, and historical scheduling data corresponding to the historical data, the historical scheduling data including scheduling decisions determined based on the historical data; inputting the historical data into an initial fusion model for training to obtain the pre-built fusion model, the initial fusion model being constructed based on an improved TS-BERT large model, an LSTM model, and an XGBoost model; wherein, the improved TS-BERT large model is obtained by adding a preset knowledge mask to the multi-head attention mechanism of the TS-BERT large model, the preset knowledge mask being used to enhance the attention weights of the target feature combination during training.

[0012] Optionally, the production data includes the output of the coal mining face, tunneling progress data, the start-up and shutdown status of the coal mining face, and the number of workers in each coal mining face; the equipment data includes the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, fault codes, cumulative running time, and daily power consumption per unit of equipment; the cost data includes the material cost per ton of coal, labor cost, transportation cost, real-time coal market price, and material purchase price; the safety data includes gas concentration, roof displacement, underground rainfall, and number of violations; and the acquisition of raw, multi-source heterogeneous data from the coal mine includes: OPC data obtained using the MES system. The UA interface collects the output, tunneling progress data, and start / stop status of the coal mining face, and uses a personnel positioning system to obtain the number of workers at the coal mining face; based on the Modbus protocol of the PLC control system, it collects the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, the fault codes, and the cumulative running time, and obtains the daily power consumption of a single device; it extracts the material cost per ton of coal, the labor cost, the transportation cost, the real-time coal market price, and the material purchase price from the MySQL database of the ERP system; it collects the gas concentration, the roof displacement, and the underground rainfall through fiber optic sensors, and uses AI recognition technology to obtain the number of violations.

[0013] Optionally, the original multi-source heterogeneous data is preprocessed to obtain multi-source heterogeneous data, including: performing a first preprocessing on the original multi-source heterogeneous data to obtain first multi-source heterogeneous data, wherein the first preprocessing includes outlier cleaning and missing value imputation; performing feature extraction on the first multi-source heterogeneous data to obtain second multi-source heterogeneous data, wherein the feature extraction includes time feature extraction, equipment health feature extraction, and data correlation feature extraction; and performing data standardization on the second multi-source heterogeneous data to obtain the multi-source heterogeneous data.

[0014] Optionally, coal mine operation scheduling based on the prediction results includes: using a directed graph model to perform correlation analysis on the production output prediction results, the equipment operating status prediction results, the cost consumption prediction results, and the safety risk prediction results to obtain correlation analysis results; determining a scheduling strategy based on the correlation analysis results, the scheduling strategy including the implementing entity, time window, resource requirements, and expected effects; and performing coal mine operation scheduling according to the scheduling strategy.

[0015] Optionally, after performing coal mine operation scheduling based on the prediction results, the method further includes: triggering a model update mechanism when a preset trigger condition is met, wherein the preset trigger condition includes at least the deviation between the prediction results and the actual results of the coal mine operation scheduling exceeding a preset deviation range; after triggering the model update mechanism, updating the preset layer number parameters of the upper layer of the large model; and updating the training data of the large model using a sliding window mechanism.

[0016] According to another aspect of this application, a prediction device for coal mine operation and scheduling data is provided, comprising: a first acquisition unit, configured to acquire raw multi-source heterogeneous data of a coal mine and preprocess the raw multi-source heterogeneous data to obtain multi-source heterogeneous data, wherein the raw multi-source heterogeneous data includes production data, equipment data, cost data, and safety data; a prediction unit, configured to input the multi-source heterogeneous data into a pre-constructed fusion model for prediction to obtain prediction results for coal mine operation and scheduling data, wherein the pre-constructed fusion model is constructed based on a large model and multiple machine models, and the prediction results include production output prediction results, equipment operating status prediction results, cost consumption prediction results, and safety risk prediction results; and a scheduling unit, configured to perform coal mine operation and scheduling based on the prediction results.

[0017] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to execute any of the aforementioned methods for predicting coal mine operation scheduling data.

[0018] This application utilizes a technical solution to acquire raw, multi-source heterogeneous data from a coal mine. This raw data is preprocessed to obtain multi-source heterogeneous data, including production data, equipment data, cost data, and safety data. The multi-source heterogeneous data is then input into a pre-constructed fusion model for prediction, yielding prediction results for coal mine operation and scheduling. This pre-constructed fusion model is built upon a large-scale model and multiple machine learning models. The prediction results include production output predictions, equipment operating status predictions, cost consumption predictions, and safety risk predictions. Coal mine operation and scheduling are then performed based on these prediction results. This application overcomes the data silo problem by integrating information from four core areas: production, equipment, cost, and safety. The pre-constructed fusion model integrates the abstraction capabilities of a large-scale model and the professional prediction advantages of multiple machine learning models, enabling accurate predictions of production output, equipment operating status, cost consumption, and safety risks. This provides a scientific basis for coal mine operation and scheduling, thus solving the problem of low prediction accuracy in existing coal mine operation and scheduling data. Attached Figure Description

[0019] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 A hardware structure block diagram of a mobile terminal for performing a prediction method for coal mine operation scheduling data according to an embodiment of this application is shown.

[0021] Figure 2 A flowchart illustrating a method for predicting coal mine operation scheduling data according to an embodiment of this application is shown.

[0022] Figure 3 A structural block diagram of a coal mine operation scheduling data prediction device provided according to an embodiment of this application is shown.

[0023] The above figures include the following reference numerals:

[0024] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0025] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0026] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0027] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] As described in the background section, the prediction of coal mine operation and scheduling data in the prior art mainly relies on traditional statistical models. These statistical models are difficult to adapt to the complex characteristics of coal mine production. In order to solve the problem of low prediction accuracy of coal mine operation and scheduling data in the prior art, the embodiments of this application provide a method for predicting coal mine operation and scheduling data, a device for predicting coal mine operation and scheduling data, and a computer-readable storage medium.

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0030] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a method of predicting coal mine operation and scheduling data according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0031] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the coal mine operation scheduling data prediction method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0032] This embodiment provides a method for predicting coal mine operation scheduling data that runs on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0033] Figure 2 This is a flowchart of a method for predicting coal mine operation and scheduling data according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0034] Step S201: Obtain the original multi-source heterogeneous data of the coal mine, and preprocess the original multi-source heterogeneous data to obtain multi-source heterogeneous data. The original multi-source heterogeneous data includes production data, equipment data, cost data and safety data.

[0035] Specifically, raw data is obtained from four key areas of coal mine production: production, equipment, cost, and safety. This data is diverse and complex, originating from different information systems or hardware devices, and varying in format and type (heterogeneity). To make this raw data suitable for subsequent predictive analysis, it needs to be preprocessed to ensure data quality and consistency, making it suitable for the input requirements of subsequent models.

[0036] Step S202: Input the above-mentioned multi-source heterogeneous data into the pre-built fusion model for prediction to obtain the prediction results of coal mine operation and scheduling data. The above-mentioned pre-built fusion model is constructed based on a large model and multiple machine models. The prediction results include production output prediction results, equipment operating status prediction results, cost consumption prediction results, and safety risk prediction results.

[0037] Specifically, the pre-built fusion model is an ensemble learning framework that combines a large model with multiple other machine learning models. It is designed to handle temporal and nonlinear multi-source data, capturing complex correlations and making predictions. These models collectively analyze multi-source heterogeneous data to generate comprehensive predictions covering coal mine production output, equipment operating status, cost consumption, and safety risks, providing data support for subsequent scheduling decisions.

[0038] Step S203: Based on the above prediction results, conduct coal mine operation scheduling.

[0039] Specifically, based on the predictions generated by pre-built fusion models, more precise operational scheduling strategies can be formulated to adapt to various potential future scenarios, including production fluctuations, equipment failures, cost changes, and safety considerations. The formulation and implementation of such strategies help improve the operational efficiency of coal mining enterprises, reduce unnecessary cost expenditures, prevent potential safety accidents, and achieve optimal resource allocation and effective adjustment of production plans.

[0040] This embodiment integrates information from four core areas: production, equipment, cost, and safety, overcoming the problem of data silos. By utilizing a pre-built fusion model, which combines the abstraction capabilities of large models with the professional predictive advantages of multiple machine models, it can accurately predict production output, equipment operating status, cost consumption, and safety risks. This provides a scientific scheduling basis for coal mine operations and solves the problem of low prediction accuracy of coal mine operation scheduling data in existing technologies.

[0041] In the specific implementation process, the aforementioned multi-source heterogeneous data is input into a pre-constructed fusion model for prediction to obtain the prediction results of coal mine operation and scheduling data. This includes: inputting the aforementioned multi-source heterogeneous data into a first large model for prediction to obtain a first prediction result; simultaneously, using the aforementioned first large model to extract features from the aforementioned multi-source heterogeneous data to obtain multi-dimensional feature vectors, which include time-series dependent features and cross-domain correlation features; inputting the aforementioned multi-dimensional feature vectors into a first machine learning model and a second machine learning model for prediction to obtain a second prediction result and a third prediction result, respectively; and determining the prediction result based on the aforementioned first prediction result, the aforementioned second prediction result, and the aforementioned third prediction result.

[0042] Specifically, heterogeneous data from multiple sources is input into a large-scale model for prediction. This model not only predicts future data but also extracts key features from the data at this stage. Through in-depth analysis of the large-scale model, a multi-dimensional feature vector is obtained, containing two types of features: time-dependent features and cross-domain correlation features. Time-dependent features refer to the correlations exhibited by data over time, such as the trend of equipment operating status over time. Cross-domain correlation features refer to the potential relationships between production, equipment, cost, and safety data, such as the correlation between equipment failure and cost increases. These multi-dimensional feature vectors are then input into two machine learning models: a first machine learning model (such as LSTM) and a second machine learning model (such as XGBoost). Each model has its unique advantages. LSTM is particularly good at processing time-series data and can provide accurate predictions of equipment status in the short term; while XGBoost excels in predicting cost data with non-linear relationships. The first and second machine learning models generate their respective prediction results, namely the second and third prediction results, which are different interpretations of a specific dataset. The final prediction result is a comprehensive consideration of the first, second, and third prediction results. This step involves weighted fusion, where the weights of each model are dynamically adjusted based on the accuracy of its predictions to ensure the reliability and accuracy of the final prediction result.

[0043] The core forecasting dimensions (covering all coal mine scheduling scenarios) include production forecasting, equipment forecasting, fault forecasting, cost forecasting, and safety forecasting. Examples of these forecasting dimensions are as follows: Regarding production forecasting, in the short term (1-7 days): based on TS-BERT+LSTM, hourly output for each working face is provided (accuracy ±5%), considering constraints such as "equipment maintenance plans" and "operator scheduling"; in the medium to long term (1-3 months): combined with XGBoost to correct for market demand impact, monthly output target suggestions are provided (e.g., "Due to rising coal prices, a 5% monthly increase in production is recommended"). Regarding equipment forecasting, this includes fault forecasting and remaining lifespan (RUL). Fault forecasting: TS-BERT extracts the coupling characteristics of "temperature-current-vibration" in the equipment, outputting the probability of failure in the next 24 hours (e.g., "The failure probability of the No. 3 coal mining machine's gearbox is 78%"); RUL prediction: based on a degradation model (e.g., Weibull distribution), combined with the equipment's cumulative runtime and fault history, "The remaining lifespan of the scraper conveyor is 1200 hours." Regarding cost forecasting, this includes the decomposition of cost per ton of coal and the cost optimization space. Cost breakdown per ton of coal: Breaking it down into "material costs (30%) + labor (25%) + energy consumption (15%) + other (30%)", and predicting fluctuations for each component (e.g., "rising steel prices lead to an 8% increase in material costs"); Cost optimization potential: Calculating quantifiable benefits such as "premature maintenance by 1 hour can reduce energy costs by 2000 yuan". Safety prediction includes risk level classification and risk diffusion simulation. Risk level classification: Based on data such as gas concentration and roof displacement, outputting three levels of warnings: red (high risk, shutdown required), yellow (medium risk, production reduced by 30%), and blue (low risk); Risk diffusion simulation: Predicting "a 60% probability that gas in a certain area will spread to adjacent working faces within 1 hour after exceeding the limit," assisting in evacuation decisions.

[0044] By inputting multi-source heterogeneous data into a highly abstract primary model, such as an improved TS-BERT, not only is an initial prediction profile obtained, but also multi-dimensional feature vectors, including those related to temporal dependencies and cross-domain associations, are extracted, enhancing the data's expressive power. Subsequently, these feature vectors are passed to refined temporal prediction models (such as LSTM) and nonlinear static data prediction models (such as XGBoost). The prediction results of these two models, together with the initial prediction, constitute a prediction system, with each dimension's prediction undergoing deep analysis by the machine learning model. Through a weighted fusion mechanism, these three prediction results are comprehensively considered, resulting in a final prediction with stronger robustness and reliability. This method not only improves prediction accuracy but also fully explores the intrinsic connections between data, ensuring the scientific and forward-looking nature of coal mine operation and scheduling decisions.

[0045] Furthermore, based on the first prediction result, the second prediction result, and the third prediction result, the prediction result is determined by: weighting and fusing the first prediction result, the second prediction result, and the third prediction result to obtain the prediction result, wherein the weights of the first prediction result, the second prediction result, and the third prediction result are adjusted based on the prediction error of the large model.

[0046] Specifically, the predictions from the first main model, the first machine learning model, and the second machine learning model—that is, the first prediction result, the second prediction result, and the third prediction result—are fused according to certain weights. These weights are not fixed but dynamic, depending on the performance of each model in the prediction process, primarily the magnitude of the prediction error. The weights are set based on the prediction error of the first main model. If a model has a smaller prediction error (more accurate prediction), then its weight in the fusion process will be higher, and vice versa. The purpose of this is to make the prediction results more favorable to those models that perform better under the current conditions, thereby improving the overall prediction accuracy.

[0047] In this embodiment, a two-layer architecture of "feature-level fusion + decision-level fusion" is adopted. The first layer is feature-level fusion (large model-driven): TS-BERT extracts features from standardized multi-source data, outputting a 128-dimensional deep feature vector, including: time-dependent features (such as the trend correlation between output and equipment energy consumption in the past 7 days); and cross-domain correlation features (such as the implicit logic of "steel price increase → material cost increase → willingness to adjust output"). The deep feature vector is simultaneously input into LSTM and XGBoost models. The second layer is decision-level fusion (dynamic weighting): the prediction results output by each model (such as output prediction values) are weighted and fused, with the weights dynamically adjusted based on real-time error. When the TS-BERT prediction error is <5%, the weight can be set to 0.6 (dominant), LSTM weight 0.3, and XGBoost weight 0.1; when there is a sudden equipment failure (error >10%), the LSTM weight is temporarily increased to 0.5 (for more sensitive short-term response), and the TS-BERT weight is reduced to 0.4.

[0048] Through the above steps, the prediction results are optimized. This not only fully integrates the strengths of different model types but also dynamically adjusts the weights based on the model's real-time performance, ensuring that the prediction results maintain high accuracy and reliability even in the complex and ever-changing coal mine operating environment. This method effectively improves the quality of prediction-based scheduling decisions, helping coal mining enterprises better plan production, maintain equipment, control costs, and mitigate safety risks.

[0049] In some embodiments of this application, before inputting the aforementioned multi-source heterogeneous data into a pre-built fusion model for prediction to obtain the prediction results of coal mine operation scheduling data, the method further includes: acquiring the historical data of the coal mine, the historical data including historical generation data, historical equipment data, historical cost data, historical safety data, and historical scheduling data corresponding to the aforementioned historical data, the aforementioned historical scheduling data including scheduling decisions determined based on the aforementioned historical data; inputting the aforementioned historical data into an initial fusion model for training to obtain the aforementioned pre-built fusion model, the aforementioned initial fusion model being constructed based on an improved TS-BERT large model, an LSTM model, and an XGBoost model; wherein, the aforementioned improved TS-BERT large model is obtained by adding a preset knowledge mask to the multi-head attention mechanism of the TS-BERT large model, the aforementioned preset knowledge mask being used to enhance the attention weights of the target feature combination during training.

[0050] Specifically, the pre-built fusion model needs to be trained and optimized using historical data before it can be officially used for prediction. First, historical data from the coal mine is acquired. This historical data includes past production data, equipment operating status, cost consumption, safety indicators, and actual scheduling decisions made based on this historical data (historical scheduling data). This constitutes a comprehensive and rich dataset, providing a solid foundation for model training.

[0051] Next, this historical data is input into the initial fusion model for training. The initial fusion model is built based on an improved TS-BERT large model, an LSTM model, and an XGBoost model. Specifically, the base large model is an improved TS-BERT (Time-Series BERT) model with an optimized structure for long-term coal mine data: encoder layers: 12 layers (adapting to long-term dependency capture of time-series data exceeding one year); attention heads: 8 heads (balancing computational efficiency and multi-feature association capture capability); input sequence length: 512 (covering high-frequency data of 30 days × 16 sampling points / day); improvements include: adding a pre-set knowledge mask to the multi-head attention mechanism. This pre-set knowledge mask is a domain knowledge mask, forcing the model to focus on key coal mine correlations such as "gas concentration-equipment operation" and "coal seam hardness-production," thus strengthening the weight of key features (such as the correlation between gas concentration and equipment operating status, or the correlation between coal seam hardness and production), thereby improving the model's learning ability and prediction accuracy for coal mine-specific situations.

[0052] The auxiliary models are the LSTM model (the first machine learning model) and the XGBoost model (the second machine learning model). The LSTM model (used for short-term refined prediction): hidden layer dimension 128, time step 24 (adapted to hourly equipment status prediction); the XGBoost model (used for non-linear static data prediction such as cost): tree depth 6, learning rate 0.1 (adapted to data affected by market fluctuations such as coal price and material cost).

[0053] Through continuous training, the initial fusion model gradually adjusts its parameters, learns patterns and rules from historical data, and eventually forms a pre-built fusion model. This process is a stage of model self-improvement and capability enhancement, ensuring that in actual predictions, the model has the ability to extract valuable information from historical experience and can more accurately predict future production, equipment, costs, and safety data.

[0054] By fully utilizing historical data, a fusion model optimized through deep learning was constructed, which not only improved prediction accuracy but also enhanced the model's understanding and adaptability to the complex environment of coal mines. By embedding specific domain knowledge masks into the TS-BERT model, attention to key feature combinations was specifically strengthened, resulting in superior performance in predicting nonlinear and multi-factor coupled tasks. This approach ensures the reliability of future prediction results and provides strong data-driven support for the operational scheduling decisions of coal mining enterprises.

[0055] In some embodiments of this application, the aforementioned production data includes the output of the coal mining face, tunneling progress data, the start-up and shutdown status of the aforementioned coal mining face, and the number of workers at each of the aforementioned coal mining faces; the aforementioned equipment data includes the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, fault codes, cumulative running time, and daily power consumption per unit of equipment; the aforementioned cost data includes the material cost per ton of coal, labor costs, transportation costs, real-time coal market price, and material purchase price; the aforementioned safety data includes gas concentration, roof displacement, underground rainfall, and the number of violations; and the acquisition of raw, multi-source heterogeneous data from the coal mine includes: OPC data obtained using a MES system. The UA interface collects the production output, tunneling progress data, and start / stop status of the coal mining face, and uses a personnel positioning system to obtain the number of workers at the coal mining face; based on the Modbus protocol of the PLC control system, it collects the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, the fault codes, and the cumulative running time, and obtains the daily power consumption of a single device; it extracts the material cost per ton of coal, the labor cost, the transportation cost, the real-time coal market price, and the material purchase price from the MySQL database of the ERP system; it collects the gas concentration, the roof displacement, and the underground rainfall through fiber optic sensors, and uses AI recognition technology to obtain the number of violations.

[0056] Specifically, regarding production data acquisition, the MES system's OPC UA interface collects real-time data on coal face output (accurate to "tons / 5 minutes"), tunneling progress (meters / hour), and the start / stop status of the coal face (Boolean value); the personnel positioning system acquires the number of workers at the coal face (persons / shift). Regarding equipment data acquisition, the PLC control system's Modbus protocol collects operating parameters (speed r / min, current A, temperature ℃), fault codes (e.g., "E102" indicating motor overload), and cumulative operating time (hours) of the coal mining machine and scraper conveyor; the energy consumption monitoring module collects daily power consumption (kWh) for each piece of equipment. Regarding cost data acquisition, the system connects to the ERP system's MySQL database to extract material costs per ton of coal (RMB / ton), labor costs (RMB / month), and transportation costs (RMB / ton). (km); real-time coal market price (yuan / ton) and steel and other material purchase price (yuan / ton) are obtained through third-party APIs. Regarding the acquisition of safety data, gas concentration (%CH4, sampling frequency 1 time / minute), roof displacement (mm / hour), and underground rainfall (mm / day) are collected through fiber optic sensors of the safety monitoring system; the number of violations (times / day) is obtained through the AI ​​recognition module of video surveillance.

[0057] The data obtained above is stored using a hybrid approach of time-series database and relational database. The time-series database (such as InfluxDB) stores high-frequency dynamic data (equipment operating parameters, gas concentration), supporting millisecond-level writes and time-range queries; the relational database (such as PostgreSQL) stores static / low-frequency data (cost data, geological parameters), ensuring efficient structured queries.

[0058] By utilizing MES, PLC control, ERP systems, and cutting-edge fiber optic sensing and AI recognition technologies, real-time monitoring and precise capture of key operational indicators are achieved, such as coal face production dynamics, equipment health status, cost fluctuations, and immediate assessment of safety risks. This not only ensures that the input data for the predictive model comprehensively reflects the real-time operational status of the coal mine, but also improves the automation and accuracy of data acquisition by adopting standardized industrial communication protocols (OPC UA, Modbus) and database interfaces (MySQL), reducing manual intervention and accelerating data flow. Furthermore, the real-time monitoring of violations using AI recognition technology further enhances the effectiveness of coal mine safety management. Overall, this embodiment ensures the quality of subsequent predictive analysis and provides solid data support for coal mine enterprise operational scheduling decisions.

[0059] In some embodiments of this application, the above-mentioned original multi-source heterogeneous data is preprocessed to obtain multi-source heterogeneous data, including: performing a first preprocessing on the above-mentioned original multi-source heterogeneous data to obtain first multi-source heterogeneous data, wherein the first preprocessing includes outlier cleaning and missing value imputation; performing feature extraction on the above-mentioned first multi-source heterogeneous data to obtain second multi-source heterogeneous data, wherein the feature extraction includes time feature extraction, equipment health feature extraction, and data correlation feature extraction; and performing data standardization processing on the above-mentioned second multi-source heterogeneous data to obtain the above-mentioned multi-source heterogeneous data.

[0060] Specifically, before forecasting coal mine operation and scheduling data, the acquired raw, multi-source, heterogeneous data needs to be preprocessed. First, outlier cleaning is performed (given the susceptibility of coal mine sensors to interference). Significant outliers are removed based on the 3σ principle: for example, a sudden increase in gas concentration to 5% (far exceeding the safety threshold of 1%) is considered sensor malfunction data. The isolated forest algorithm identifies latent anomalies: such as contradictory data like a sudden drop in operating current but unchanged output (possibly due to poor sensor contact). A manual verification mechanism is implemented: manual review nodes are set up for critical safety data (such as roof displacement exceeding 20mm) to avoid mistakenly deleting valid early warning information. Next, missing value imputation is performed (addressing data disconnection caused by unstable underground networks). For short-term missing data (<1 hour): linear interpolation is used (e.g., equipment temperature data); for long-term missing data (>1 hour): the average of data from similar equipment is used for imputation (e.g., when a coal mining machine malfunctions, the average energy consumption of the same model is used as a substitute); special handling of safety data: when gas concentration is missing, weighted imputation based on neighboring sensor data is automatically triggered (weight is inversely proportional to distance) to ensure continuous safety monitoring. After outlier cleaning and missing value imputation, the first multi-source heterogeneous data was obtained.

[0061] Feature extraction, or feature engineering, was performed on the first multi-source heterogeneous data (extracting key features of the coal mine scenario). Temporal features included: extracting "weekdays / holidays," "quarterly cycles" (e.g., a 10% decrease in winter production due to anti-freezing measures), and "maintenance windows" (e.g., weekly maintenance shutdowns from 3-6 AM on Wednesdays); equipment health features included: calculating "failure frequency = monthly failure count / operating time" and "energy consumption fluctuation rate = (daily energy consumption - average) / average"; and coupling features (correlation features between data points): constructing a "geological conditions - production" correlation feature (e.g., production decreases by 8% for every 1-level increase in coal seam hardness) and a "gas concentration - equipment load" limiting feature (equipment load is forcibly reduced to 70% when gas exceeds 0.8%). By extracting features from the first multi-source heterogeneous data, the second multi-source heterogeneous data was obtained.

[0062] The second type of multi-source heterogeneous data is subjected to data standardization (eliminating the influence of dimensions). Linear standardization: For continuous data (such as output, cost), the Min-Max transformation is used to transform it to the interval [0,1], the formula is: x′=max(x)-min(x)x-min(x); Normalization: For discrete data (such as fault codes, work surface status), one-hot encoding is used, such as "E102 fault" is encoded as [1,0,0], and "normal" is encoded as [0,1,0].

[0063] The aforementioned preprocessing steps significantly improve the quality of the input data, providing a clean, standardized, and highly characteristic dataset for model training and prediction. This is a crucial prerequisite for ensuring excellent predictive model performance and reliable prediction results. Through this series of preprocessing steps, the complex and diverse raw data is transformed into a clear, coherent, and information-rich format, enabling subsequent large-scale model fusion predictions to more accurately reflect the actual operating conditions of the coal mine, thereby providing strong support for efficient operational scheduling decisions.

[0064] In some embodiments of this application, coal mine operation scheduling based on the above prediction results includes: using a directed graph model to perform correlation analysis on the above production output prediction results, the above equipment operating status prediction results, the above cost consumption prediction results, and the above safety risk prediction results to obtain correlation analysis results; determining a scheduling strategy based on the above correlation analysis results, wherein the scheduling strategy includes the executing entity, time window, resource demand, and expected effect; and performing the above coal mine operation scheduling according to the above scheduling strategy.

[0065] Specifically, by constructing a directed graph model, intuitive causal relationships are established between data from different dimensions, such as predicted production output, equipment operating status, cost consumption, and safety risks. For example, a decrease in output caused by equipment failure is directly mapped onto the graph. The graph model also shows the potential cost increases and changes in safety risks that this decrease may cause, enabling schedulers to quickly understand the interplay between various predicted outcomes. This method significantly improves the transparency and comprehensiveness of scheduling decisions, ensuring that decision-makers can make optimal scheduling decisions based on all relevant factors. Based on the results of the above correlation analysis, a detailed scheduling strategy is determined. This strategy clearly specifies the implementing entity (i.e., which department or individual is responsible for execution), the time window (optimal execution time), resource requirements (including the required manpower, materials, and equipment), and expected results (the anticipated benefits or improvements after implementing the strategy). The generation process of this strategy considers both the predicted results and historical best practices, as well as the actual constraints of the coal mine, ensuring the feasibility and effectiveness of the scheduling recommendations. Adjustments are then made to the coal mine's operational scheduling based on the determined scheduling strategy.

[0066] Example as follows:

[0067] 1. Equipment failure leads to production loss, which in turn increases costs: If "No. 2 coal mining machine malfunctions and shuts down for 4 hours" is predicted, the following will be automatically triggered:

[0068] Production loss calculation: 4 hours × average capacity of 150 tons / hour = 600 tons;

[0069] Cost impact: 600 tons × (profit of 300 yuan per ton of coal) = loss of 180,000 yuan;

[0070] Response plan simulation: "Using the backup coal mining machine (which requires 2 hours to start) can reduce losses by 300 tons, with an additional cost of 5,000 yuan."

[0071] 2. Safety risks lead to production adjustments, which in turn lead to cost compensation: If it is predicted that "gas concentration will exceed 1% (red alert)," the system will automatically analyze the following:

[0072] Production adjustment: The working face needs to be shut down for 2 hours for ventilation, which will affect the output by 300 tons;

[0073] Cost compensation: 200 tons will be recovered through "1 hour of overtime work on other work sites", with an additional labor cost of 20,000 yuan;

[0074] Net impact: Final loss of 100 tons (30,000 yuan), but safety accident avoided (potential loss of more than 1 million yuan).

[0075] By closely linking forecasting results with scheduling decisions, a complete closed loop is formed from forecasting to decision-making to execution. This not only improves the scientific rigor and relevance of scheduling decisions but also significantly enhances the efficiency and effectiveness of scheduling execution. It provides coal mining enterprises with intelligent and refined operational scheduling tools, helping them to optimize resource allocation, effectively control costs, and promptly mitigate safety risks.

[0076] In some embodiments of this application, the scheduling strategy generation logic is as follows: a strategy (suggestion) is generated based on "prediction results + constraints + historical best solutions," for example, including:

[0077] Implementing entity: Clearly define the responsible department (e.g., "The Mechanical and Electrical Department is responsible for the maintenance of Coal Mining Machine No. 3").

[0078] Time window: Provides the optimal execution time (e.g., "Avoid the 8:00 AM production peak, maintenance is recommended between 9:00 AM and 11:00 AM").

[0079] Resource requirements: Related materials, personnel, equipment, etc. (e.g., "2 electricians and 1 set of gearbox spare parts are needed").

[0080] Expected results: Quantifiable benefits (e.g., "After maintenance, the probability of failure can be reduced to 12%, ensuring that this week's production targets are met").

[0081] Some embodiments of this application also include a visual interactive design (adapted to the dispatch center's large screen). The core dashboard module includes: a real-time prediction curve: comparing the deviation between "predicted output" and "planned output" (red line represents prediction, blue line represents plan); an equipment health heatmap: using color depth to indicate the probability of failure for each piece of equipment (red ≥80%, yellow 50%-80%); a risk diffusion dynamic map: simulating the spatial diffusion path of gas and roof risks (updated every 10 minutes); and suggestion execution tracking: displaying the status of dispatch suggestions as "accepted," "in progress," or "completed," and associating them with the responsible person. Interactive functions include: support for drill-down queries (e.g., clicking "cost overrun" to view the specific influencing factors of "steel price increase") and historical backtracking (comparing the prediction accuracy of the past three similar scenarios).

[0082] In some embodiments of this application, after coal mine operation scheduling is performed based on the above prediction results, the method further includes: triggering a model update mechanism when a preset trigger condition is met, wherein the preset trigger condition includes at least the deviation between the above prediction results and the actual results of the above coal mine operation scheduling exceeding a preset deviation range; after triggering the above model update mechanism, updating the preset layer number parameters of the upper layer of the above large model; and updating the training data of the above large model using a sliding window mechanism.

[0083] Specifically, this embodiment introduces a deviation monitoring mechanism. This mechanism periodically compares the predicted results with the actual execution results of coal mine operation and scheduling. If the deviation between the two exceeds a pre-set range, the model update process is automatically triggered. This real-time feedback mechanism allows for timely adjustment of the prediction model based on actual conditions, avoiding prediction inaccuracies due to model obsolescence or environmental changes, thus enhancing the robustness and adaptability of the prediction model. After triggering the model update mechanism, the focus is on updating the preset parameters of the upper layers of the larger model, rather than the entire model. This partial update strategy considers both the model's need to adapt to new data trends and avoids unnecessary perturbations to the stable parts of the model that are suitable for long-term data patterns, maintaining the model's stability and generalization ability. Updating the upper-layer parameters can more quickly reflect changes in recent data characteristics, improving the model's sensitivity to short-term environmental changes. To ensure the timeliness and representativeness of the model training dataset, a sliding window mechanism is used for data updates. During each model update, the latest data is added to the end of the training set, while the oldest data is removed, thus ensuring that the training set always covers data within a certain time window. This method effectively solves the problem of lag in traditional models when facing dynamic environmental changes, enabling the model to continuously learn and adapt to the latest production conditions, thus maintaining the vitality and accuracy of the predictive model.

[0084] Specifically, preset trigger conditions can also be based on event triggering and periodic triggering. An example of the quantitative design of preset trigger conditions is as follows:

[0085] Data deviation trigger: The deviation between the predicted value and the actual value of a certain dimension exceeds the threshold (output ±8%, equipment failure probability ±15%).

[0086] Event triggers: Equipment reports a fault code (e.g., "E201" indicates a gearbox malfunction), methane concentration exceeds 0.8%, or daily coal price fluctuation exceeds 10%;

[0087] Periodic trigger: Automatically execute full data training at 2 AM every day (outside of peak production times, to reduce system load).

[0088] An example of the incremental training process is as follows: Freeze the parameters of the bottom 6 layers of TS-BERT (preserving the ability to extract general temporal features) and only update the parameters of the top 6 layers; adopt a sliding window mechanism (window size = 3 months of data) to replace the oldest data with the latest data to avoid the model "forgetting" historical patterns; control the training time within 30 minutes (compress the amount of computation through model distillation technology) to ensure that it does not affect production scheduling.

[0089] When the deviation between the predicted result and the actual scheduling data exceeds a preset threshold, an incremental update of the model is triggered. This involves fine-tuning the parameters of only a preset number of upper layers (e.g., the last 6 layers) of the large model. This preserves the general feature extraction capabilities of the lower layers while quickly adapting to recent scene changes, avoiding "catastrophic forgetting." Simultaneously, a sliding window mechanism dynamically replaces old data and retains the most recent high-quality samples, ensuring that the training data always reflects current production patterns, balancing stability and timeliness. This mechanism enables the model to continuously evolve without manual intervention, significantly reducing the risk of model drift and ensuring long-term stability of prediction accuracy.

[0090] In this application, after integrating a large model, the error rate for production output and equipment failure prediction is reduced to below 8%, a 40% decrease compared to traditional models. This can reduce monthly production loss by more than 500 tons and reduce cost waste by 15%. The model's incremental training cycle is 30 minutes per cycle, and the prediction results are updated within one hour after equipment failure, providing minute-level support for scheduling and preventing the escalation of failures. It not only outputs prediction data but also generates specific scheduling actions (maintenance time, material allocation quantity), improving scheduling efficiency by 30% and solving the problem of "useless predictions." The core architecture is compatible with both underground and open-pit coal mines; only the data input dimensions need to be adjusted (e.g., adding a "stripping quantity" parameter for open-pit coal mines), reducing enterprise implementation costs by 20%. Safety predictions provide early warnings 6-8 hours in advance, 4 hours earlier than traditional manual inspections, reducing underground safety hazards.

[0091] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the prediction method for coal mine operation scheduling data of this application will be described in detail below with reference to specific embodiments.

[0092] This embodiment relates to a specific method for predicting coal mine operation and scheduling data. Taking "dynamic monthly output prediction of a large underground coal mine" as an example, the implementation steps and cycle are as follows to ensure that the solution is feasible and reproducible:

[0093] 1. Data Acquisition Phase (Days 1-2): Collect core data from the coal mine over the past 3 years (36 months): monthly output, number of coal faces, equipment operating time, coal seam thickness / hardness, monthly rainfall (affecting underground operations), and number of workers. Access the above data in real time through the coal mine's existing data platform API interface to form an initial dataset (36 records, 12 feature dimensions).

[0094] 2. Data Preprocessing Stage (Day 3): Data cleaning using Python Pandas library: Remove 2 sensor fault data points with "output of 0", and fill in 3 missing coal seam thickness data points using linear interpolation. Feature extraction: Identify the equipment's "72-hour maintenance cycle" and label the associated feature "output decreases by 10% when rainfall exceeds 50mm". Standardization: Use the Min-Max method to convert "equipment operating time (0-720 hours)" and "coal seam thickness (1-5 meters)" into the [0,1] interval to generate a standardized dataset.

[0095] 3. Model Building and Training Phase (Days 4-5): A Time-Series BERT+LSTM fusion model was built based on the PyTorch framework, with the following core parameters set: Training epochs: 50; Batch size: 8; Learning rate: 1e-4; Feature input dimension: 12. The training set (25 months) and test set (11 months) were divided in a 7:3 ratio. After training, the model's output prediction error rate on the test set was 6.2%, meeting the enterprise's requirement of "error ≤ 10%".

[0096] 4. Prediction and Scheduling Application Phase (Day 6 and beyond): The latest equipment / geological / personnel data is automatically collected daily at 8:00 AM, input into the model to generate the next day's production forecast, and summarized into a monthly forecast report. When the model predicts that weekly production will be 5% lower than planned (planned 15,000 tons, predicted 14,250 tons), a scheduling suggestion is automatically pushed: "Increase the working hours of working face No. 2 by 2 hours / day to make up for the production shortfall of 750 tons; 5 workers need to work overtime simultaneously." This suggestion is simultaneously displayed on the dispatch center's large screen to track its implementation status.

[0097] This application uses a random forest model instead of a large model to predict multi-source data, filtering key factors such as "equipment runtime" and "coal seam thickness" based on feature importance. The advantages are low training cost (no high-performance GPU required) and simple deployment; the disadvantages are weak long-term data processing capability, prediction errors exceeding 15%, and inability to capture implicit correlations. It is suitable for scenarios involving small coal mines, data volumes less than one year, and limited budgets.

[0098] This paper replaces the Time-Series BERT model with a CNN-LSTM model. CNN extracts spatial features of equipment (such as the correlation of equipment status across different working faces), while LSTM extracts temporal features, thus constructing a CNN-LSTM fusion model. The advantages are strong spatial feature extraction capabilities and adaptability to coal mines with complex equipment distribution. The disadvantages are insufficient capture of long-term (over 6 months) data dependencies and a 30% lower prediction timeliness. It is suitable for large underground coal mines with complex underground equipment distribution where spatial correlation analysis is crucial.

[0099] This approach uses a single LSTM model instead of a fusion architecture, processing multi-source data with only the LSTM model and avoiding the introduction of a large model. Its advantages include a simple model structure and short training cycle (within 10 hours). Disadvantages include weak deep feature extraction capabilities and prediction errors exceeding 12% in multi-factor coupled scenarios. The application scenario is coal mines where prediction accuracy requirements are not high (error ≤ 15%) and the production scenario is relatively simple.

[0100] This application also provides a device for predicting coal mine operation and scheduling data. It should be noted that this device can be used to execute the method for predicting coal mine operation and scheduling data provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0101] The following describes the prediction device for coal mine operation and scheduling data provided in the embodiments of this application.

[0102] Figure 3 This is a structural block diagram of a coal mine operation scheduling data prediction device according to an embodiment of this application. Figure 3As shown, the device includes a first acquisition unit 10, a prediction unit 20, and a scheduling unit 30. The first acquisition unit is used to acquire raw multi-source heterogeneous data of the coal mine and preprocess the raw multi-source heterogeneous data to obtain multi-source heterogeneous data, which includes production data, equipment data, cost data, and safety data. The prediction unit is used to input the multi-source heterogeneous data into a pre-built fusion model for prediction to obtain the prediction results of coal mine operation scheduling data. The pre-built fusion model is constructed based on a large model and multiple machine models. The prediction results include production output prediction results, equipment operating status prediction results, cost consumption prediction results, and safety risk prediction results. The scheduling unit is used to perform coal mine operation scheduling based on the above prediction results.

[0103] This embodiment integrates information from four core areas: production, equipment, cost, and safety, overcoming the problem of data silos. By utilizing a pre-built fusion model, which combines the abstraction capabilities of large models with the professional predictive advantages of multiple machine models, it can accurately predict production output, equipment operating status, cost consumption, and safety risks. This provides a scientific scheduling basis for coal mine operations and solves the problem of low prediction accuracy of coal mine operation scheduling data in existing technologies.

[0104] In the specific implementation process, the prediction unit includes a first prediction module, a second prediction module, and a first determination module. The first prediction module is used to input the multi-source heterogeneous data into a first large model for prediction to obtain a first prediction result. At the same time, the first large model is used to extract features from the multi-source heterogeneous data to obtain a multi-dimensional feature vector, which includes time-series dependent features and cross-domain correlation features. The second prediction module is used to input the multi-dimensional feature vector into a first machine learning model and a second machine learning model for prediction to obtain a second prediction result and a third prediction result, respectively. The first determination module is used to determine the prediction result based on the first prediction result, the second prediction result, and the third prediction result.

[0105] By inputting multi-source heterogeneous data into a highly abstract primary model, such as an improved TS-BERT, not only is an initial prediction profile obtained, but also multi-dimensional feature vectors, including those related to temporal dependencies and cross-domain associations, are extracted, enhancing the data's expressive power. Subsequently, these feature vectors are passed to refined temporal prediction models (such as LSTM) and nonlinear static data prediction models (such as XGBoost). The prediction results of these two models, together with the initial prediction, constitute a prediction system, with each dimension's prediction undergoing deep analysis by the machine learning model. Through a weighted fusion mechanism, these three prediction results are comprehensively considered, resulting in a final prediction with stronger robustness and reliability. This method not only improves prediction accuracy but also fully explores the intrinsic connections between data, ensuring the scientific and forward-looking nature of coal mine operation and scheduling decisions.

[0106] Furthermore, the first determining module includes a weighted fusion submodule, which is used to perform weighted fusion on the first prediction result, the second prediction result and the third prediction result to obtain the prediction result. The weights of the first prediction result, the second prediction result and the third prediction result are adjusted based on the prediction error of the large model.

[0107] This embodiment optimizes the prediction results by fully combining the strengths of different model types and dynamically adjusting weights based on the model's real-time performance. This ensures that the prediction results maintain high accuracy and reliability even in the complex and ever-changing coal mine operating environment. This method effectively improves the quality of prediction-based scheduling decisions, helping coal mining enterprises better plan production, maintain equipment, control costs, and mitigate safety risks.

[0108] In some embodiments of this application, the above-mentioned apparatus further includes a second acquisition unit and a training unit. The second acquisition unit is used to acquire historical data of the coal mine before inputting the multi-source heterogeneous data into a pre-built fusion model for prediction to obtain the prediction results of the coal mine operation scheduling data. The historical data includes historical generation data, historical equipment data, historical cost data, historical safety data, and historical scheduling data corresponding to the historical data. The historical scheduling data includes scheduling decisions determined based on the historical data. The training unit is used to input the historical data into an initial fusion model for training to obtain the pre-built fusion model. The initial fusion model is constructed based on an improved TS-BERT large model, an LSTM model, and an XGBoost model. The improved TS-BERT large model is obtained by adding a preset knowledge mask to the multi-head attention mechanism of the TS-BERT large model. The preset knowledge mask is used to enhance the attention weights of the target feature combination during training.

[0109] By fully utilizing historical data, a fusion model optimized through deep learning was constructed, which not only improved prediction accuracy but also enhanced the model's understanding and adaptability to the complex environment of coal mines. By embedding specific domain knowledge masks into the TS-BERT model, attention to key feature combinations was specifically strengthened, resulting in superior performance in predicting nonlinear and multi-factor coupled tasks. This approach ensures the reliability of future prediction results and provides strong data-driven support for the operational scheduling decisions of coal mining enterprises.

[0110] In some embodiments of this application, the aforementioned production data includes the output of the coal mining face, tunneling progress data, the start-up and shutdown status of the aforementioned coal mining face, and the number of workers at each of the aforementioned coal mining faces; the aforementioned equipment data includes the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, fault codes, cumulative running time, and daily power consumption per unit; the aforementioned cost data includes the material cost per ton of coal, labor cost, transportation cost, real-time coal market price, and material purchase price; the aforementioned safety data includes gas concentration, roof displacement, underground rainfall, and number of violations; and the aforementioned first acquisition unit includes a first acquisition module, a second acquisition module, a third acquisition module, and a fourth acquisition module. The first acquisition module is used to collect the production output, tunneling progress data, and start / stop status of the coal mining face using the OPC UA interface of the MES system, and to obtain the number of workers at the coal mining face using a personnel positioning system; the second acquisition module is used to collect the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, the fault codes, and the cumulative running time based on the Modbus protocol of the PLC control system, and to obtain the daily power consumption of a single device; the third acquisition module is used to extract the material cost per ton of coal, the labor cost, the transportation cost, the real-time coal market price, and the material purchase price from the MySQL database of the ERP system; the fourth acquisition module is used to collect the gas concentration, the roof displacement, and the underground rainfall through fiber optic sensors, and to obtain the number of violations through AI recognition technology.

[0111] By utilizing MES, PLC control, ERP systems, and cutting-edge fiber optic sensing and AI recognition technologies, real-time monitoring and precise capture of key operational indicators are achieved, such as coal face production dynamics, equipment health status, cost fluctuations, and immediate assessment of safety risks. This not only ensures that the input data for the predictive model comprehensively reflects the real-time operational status of the coal mine, but also improves the automation and accuracy of data acquisition by adopting standardized industrial communication protocols (OPC UA, Modbus) and database interfaces (MySQL), reducing manual intervention and accelerating data flow. Furthermore, the real-time monitoring of violations using AI recognition technology further enhances the effectiveness of coal mine safety management. Overall, this embodiment ensures the quality of subsequent predictive analysis and provides solid data support for coal mine enterprise operational scheduling decisions.

[0112] In some embodiments of this application, the first acquisition unit includes a first processing module, a second processing module, and a third processing module. The first processing module is used to perform a first preprocessing on the original multi-source heterogeneous data to obtain first multi-source heterogeneous data, wherein the first preprocessing includes outlier cleaning and missing value imputation; the second processing module is used to perform feature extraction on the first multi-source heterogeneous data to obtain second multi-source heterogeneous data, wherein the feature extraction includes time feature extraction, equipment health feature extraction, and data correlation feature extraction; the third processing module is used to perform data standardization processing on the second multi-source heterogeneous data to obtain the multi-source heterogeneous data.

[0113] The aforementioned preprocessing steps significantly improve the quality of the input data, providing a clean, standardized, and highly characteristic dataset for model training and prediction. This is a crucial prerequisite for ensuring excellent predictive model performance and reliable prediction results. Through this series of preprocessing steps, the complex and diverse raw data is transformed into a clear, coherent, and information-rich format, enabling subsequent large-scale model fusion predictions to more accurately reflect the actual operating conditions of the coal mine, thereby providing strong support for efficient operational scheduling decisions.

[0114] In some embodiments of this application, the scheduling unit includes a parsing module, a second determining module, and a scheduling module. The parsing module uses a directed graph model to correlate the production output forecast, equipment operating status forecast, cost consumption forecast, and safety risk forecast to obtain a correlation parsing result. The second determining module determines a scheduling strategy based on the correlation parsing result, the scheduling strategy including the executing entity, time window, resource requirements, and expected results. The scheduling module performs the coal mine operation scheduling according to the scheduling strategy.

[0115] By closely linking forecasting results with scheduling decisions, a complete closed loop is formed from forecasting to decision-making to execution. This not only improves the scientific rigor and relevance of scheduling decisions but also significantly enhances the efficiency and effectiveness of scheduling execution. It provides coal mining enterprises with intelligent and refined operational scheduling tools, helping them to optimize resource allocation, effectively control costs, and promptly mitigate safety risks.

[0116] In some embodiments of this application, the above-described apparatus further includes a triggering unit, a first updating unit, and a second updating unit. The triggering unit is used to trigger a model update mechanism after coal mine operation scheduling based on the above-described prediction results, provided that preset triggering conditions are met. The preset triggering conditions include at least that the deviation between the predicted results and the actual results of the coal mine operation scheduling exceeds a preset deviation range. The first updating unit is used to update the preset layer number parameters of the upper layer of the large model after the model update mechanism is triggered. The second updating unit is used to update the training data of the large model using a sliding window mechanism.

[0117] When the deviation between the predicted result and the actual scheduling data exceeds a preset threshold, an incremental update of the model is triggered. This involves fine-tuning the parameters of only a preset number of upper layers (e.g., the last 6 layers) of the large model. This preserves the general feature extraction capabilities of the lower layers while quickly adapting to recent scene changes, avoiding "catastrophic forgetting." Simultaneously, a sliding window mechanism dynamically replaces old data and retains the most recent high-quality samples, ensuring that the training data always reflects current production patterns, balancing stability and timeliness. This mechanism enables the model to continuously evolve without manual intervention, significantly reducing the risk of model drift and ensuring long-term stability of prediction accuracy.

[0118] The aforementioned coal mine operation and scheduling data prediction device includes a processor and a memory. The first acquisition unit, prediction unit, scheduling unit, etc., are all stored as program units in the memory, and the processor executes the program units stored in the memory to achieve the corresponding functions. All of the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0119] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0120] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform a method for predicting coal mine operation and scheduling data.

[0121] This invention provides a processor for running a program, wherein the program executes the prediction method for coal mine operation and scheduling data.

[0122] This invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the aforementioned method for predicting coal mine operation and scheduling data. The device described herein can be a server, PC, tablet, mobile phone, etc.

[0123] This application also provides a computer program product that, when executed on a data processing device, is adapted to perform the steps of initializing the prediction method with the aforementioned coal mine operation scheduling data.

[0124] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0125] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0126] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0127] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0128] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0129] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0130] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0131] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0132] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0133] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0134] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for predicting coal mine operation scheduling data, characterized in that, include: The raw multi-source heterogeneous data of the coal mine is acquired, and the raw multi-source heterogeneous data is preprocessed to obtain multi-source heterogeneous data, which includes production data, equipment data, cost data and safety data. The multi-source heterogeneous data is input into a pre-built fusion model for prediction to obtain the prediction results of coal mine operation and scheduling data. The pre-built fusion model is constructed based on a large model and multiple machine models. The prediction results include production output prediction results, equipment operating status prediction results, cost consumption prediction results, and safety risk prediction results. Coal mine operation scheduling is carried out based on the prediction results.

2. The method of claim 1, wherein, The multi-source heterogeneous data is input into a pre-built fusion model for prediction, resulting in predictions of coal mine operation and scheduling data, including: The multi-source heterogeneous data is input into the first large model for prediction to obtain the first prediction result. At the same time, the first large model is used to extract features from the multi-source heterogeneous data to obtain a multi-dimensional feature vector, which includes time-dependent features and cross-domain correlation features. The multidimensional feature vectors are input into the first machine learning model and the second machine learning model for prediction, respectively, to obtain the second prediction result and the third prediction result. The prediction result is determined based on the first prediction result, the second prediction result, and the third prediction result.

3. The method according to claim 2, characterized in that, Determining the prediction result based on the first prediction result, the second prediction result, and the third prediction result includes: The first prediction result, the second prediction result, and the third prediction result are weighted and fused to obtain the prediction result. The weights of the first prediction result, the second prediction result, and the third prediction result are adjusted based on the prediction error of the large model.

4. The method according to claim 1, characterized in that, Before inputting the multi-source heterogeneous data into a pre-constructed fusion model for prediction to obtain the prediction results of coal mine operation and scheduling data, the method further includes: The historical data of the coal mine is obtained, including historical generation data, historical equipment data, historical cost data, historical safety data, and historical scheduling data corresponding to the historical data. The historical scheduling data includes scheduling decisions determined based on the historical data. The historical data is input into the initial fusion model for training to obtain the pre-built fusion model, which is constructed based on the improved TS-BERT large model, LSTM model and XGBoost model; The improved TS-BERT large model is obtained by adding a preset knowledge mask to the multi-head attention mechanism of the TS-BERT large model. The preset knowledge mask is used to enhance the attention weight of the target feature combination during training.

5. The method according to claim 1, characterized in that, The production data includes the output of the coal mining face, tunneling progress data, the start / stop status of the coal mining face, and the number of workers at each coal mining face. The equipment data includes the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, fault codes, cumulative running time, and daily power consumption per unit of equipment. The cost data includes the material cost per ton of coal, labor cost, transportation cost, real-time coal market price, and material purchase price. The safety data includes gas concentration, roof displacement, underground rainfall, and the number of violations. The original multi-source heterogeneous data of the coal mine is obtained, including: The production output, tunneling progress data, and start / stop status of the coal mining face are collected using the OPC UA interface of the MES system, and the number of workers at the coal mining face is obtained using a personnel positioning system. The Modbus protocol based on the PLC control system is used to collect the operating parameters of the coal mining machine, the operating parameters of the scraper conveyor, the fault codes and the cumulative running time, and to obtain the daily power consumption of a single device. Extract the material cost per ton of coal, the labor cost, the transportation cost, the real-time coal market price, and the material purchase price from the MySQL database of the ERP system; The gas concentration, roof displacement, and downhole rainfall are collected using fiber optic sensors, and the number of violations is obtained using AI recognition technology.

6. The method according to claim 1, characterized in that, The original multi-source heterogeneous data is preprocessed to obtain multi-source heterogeneous data, including: The original multi-source heterogeneous data is subjected to a first preprocessing to obtain the first multi-source heterogeneous data. The first preprocessing includes outlier cleaning and missing value imputation. Feature extraction is performed on the first multi-source heterogeneous data to obtain the second multi-source heterogeneous data. The feature extraction includes time feature extraction, device health feature extraction, and data correlation feature extraction. The second multi-source heterogeneous data is subjected to data standardization processing to obtain the multi-source heterogeneous data.

7. The method according to claim 1, characterized in that, Coal mine operation scheduling based on the prediction results includes: A directed graph model is used to perform correlation analysis on the production output prediction results, equipment operating status prediction results, cost consumption prediction results, and safety risk prediction results to obtain correlation analysis results; The scheduling strategy is determined based on the correlation parsing results. The scheduling strategy includes the execution subject, time window, resource requirements, and expected results. The coal mine operation scheduling is carried out according to the aforementioned scheduling strategy.

8. The method according to claim 1, characterized in that, After scheduling coal mine operations based on the prediction results, the method further includes: Under the condition that a preset triggering condition is met, the model update mechanism is triggered. The preset triggering condition includes at least the deviation between the prediction result and the actual result of the coal mine operation and scheduling exceeding a preset deviation range. After the model update mechanism is triggered, the preset layer number parameter of the upper layer of the large model is updated; The training data of the large model is updated using a sliding window mechanism.

9. A device for predicting coal mine operation and scheduling data, characterized in that, include: The first acquisition unit is used to acquire the original multi-source heterogeneous data of the coal mine and preprocess the original multi-source heterogeneous data to obtain multi-source heterogeneous data. The original multi-source heterogeneous data includes production data, equipment data, cost data and safety data. The prediction unit is used to input the multi-source heterogeneous data into a pre-built fusion model for prediction, and obtain the prediction results of coal mine operation and scheduling data. The pre-built fusion model is constructed based on a large model and multiple machine models. The prediction results include production output prediction results, equipment operating status prediction results, cost consumption prediction results, and safety risk prediction results. The scheduling unit is used to schedule coal mine operations based on the prediction results.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method for predicting coal mine operation and scheduling data as described in any one of claims 1 to 8.

Images