Coal mine driving face gas concentration prediction method based on combination of theoretical model and deep learning
By combining the gas diffusion equation and the deep learning model in a dynamic fusion method, the problems of real-time performance and accuracy in predicting gas concentration in coal mine tunneling faces were solved, achieving accurate prediction of gas concentration, reducing the risk of gas disasters, and ensuring safe production in the mine.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA COAL TECH & ENG GRP CHONGQING RES INST CO LTD
- Filing Date
- 2025-12-25
- Publication Date
- 2026-04-28
AI Technical Summary
Existing gas concentration prediction methods in coal mine tunneling faces suffer from slow response and insufficient generalization ability. In particular, the prediction accuracy decreases when faced with sudden changes in production conditions. Furthermore, the theoretical models have weak response capabilities to sudden gas outbursts and nonlinear disturbances. Existing combined methods lack effective dynamic fusion mechanisms and adaptive correction strategies, resulting in large deviations between prediction results and actual working conditions.
By combining the theoretical model of the gas diffusion equation with deep learning models such as Convolutional Neural Network (CNN) and Long Short-Term Memory Network (LSTM), multi-source data is fused in real time through dynamic weighting and adaptive feedback mechanisms. The model weights are dynamically adjusted and the parameters are optimized to improve prediction accuracy.
It enables accurate and real-time prediction of gas concentration at coal mine tunneling faces, improving the accuracy and reliability of prediction, reducing the risk of gas disasters, and providing reliable protection for safe production in mines.
Smart Images
Figure CN121938501A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of coal mine gas monitoring and disaster early warning technology, and relates to a method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning. Background Technology
[0002] Accurate prediction of methane concentration at coal mine tunneling faces is crucial for ensuring miners' safety and preventing methane explosions. Currently, methane concentration prediction methods are mainly divided into two categories: one is based on data-driven machine learning models, and the other is based on theoretical models of physical mechanisms.
[0003] Data-driven machine learning models primarily rely on historical monitoring data for training, enabling them to capture the nonlinear characteristics of gas concentration changes. However, in the complex and dynamic environment of coal mine tunneling faces, gas outbursts are influenced by various factors, including sudden disturbances such as temperature, atmospheric pressure, airflow, instantaneous productivity, and coal face gas desorption. These models often suffer from slow response and insufficient generalization ability, especially when faced with abrupt changes in production conditions, resulting in a significant decrease in prediction accuracy.
[0004] Theoretical models are typically built upon gas diffusion equations and fluid dynamics principles, and can effectively describe the variation of gas concentration under ideal conditions. These models possess clear physical meaning and strong interpretability. However, their predictive accuracy heavily relies on accurate estimations of parameters such as gas emission rate and ventilation volume. In actual mine environments, these parameters are difficult to obtain accurately in real time, and the models have weak response capabilities to sudden gas outbursts and nonlinear disturbances, leading to significant deviations between predicted results and actual operating conditions.
[0005] While some existing technologies have attempted to combine machine learning with theoretical models, most remain at the stage of simple serial or parallel use, lacking effective dynamic fusion mechanisms and adaptive correction strategies. They fail to fully balance the constraints of physical laws with the nonlinear capture capabilities driven by data. Therefore, existing gas concentration prediction methods still have significant shortcomings in terms of real-time performance, accuracy, and stability, making it difficult to meet the demand for accurate early warning of gas disasters in high-gas mine tunneling operations.
[0006] In view of this, it is necessary to propose a new method for predicting gas concentration. This method can organically integrate the advantages of theoretical models and deep learning models, and achieve accurate real-time prediction of gas concentration in coal mine tunneling faces through dynamic weighting and adaptive feedback mechanisms, thereby providing more reliable technical support for safe mine production. Summary of the Invention
[0007] In view of this, the purpose of this invention is to provide a method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning.
[0008] To achieve the above objectives, the present invention provides the following technical solution: A method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning includes the following steps: S1: Collect multi-source data such as methane concentration, temperature, absolute atmospheric pressure, air volume and instantaneous productivity at the coal mine tunneling face, and transmit the data to the data processing center in real time through a sensor network; S2: Based on the gas diffusion equation, a theoretical model is constructed, and the gas concentration change rate is calculated using the temperature, absolute atmospheric pressure, air volume and gas emission rate, so as to preliminarily predict the gas concentration in the next minute. S3: The collected real-time multi-source data is input into a deep learning model that combines a convolutional neural network (CNN) and a long short-term memory network (LSTM). The CNN extracts local features from the multi-source data, and the LSTM captures long-term and short-term dependencies, thereby predicting the gas concentration within the next minute. S4: Dynamically adjust the weights of the theoretical model and the deep learning model based on historical prediction errors, and perform weighted fusion of the prediction results of the theoretical model and the prediction results of the deep learning model to obtain the final gas concentration prediction value. S5: Through an adaptive feedback mechanism, the final predicted gas concentration is compared with the actual monitored gas concentration, the prediction error is calculated, and the parameters of the deep learning model are dynamically adjusted according to the prediction error.
[0009] Furthermore, in S2, the gas diffusion equation is:
[0010] in, for t The gas concentration at any given time, for t Total gas emission at any given moment for t The air volume supplied by the local fan at any given time. V The volume of the tunneling face is monitored.
[0011] Furthermore, in S2, starting from the gas concentration at the current moment, assuming that the total gas outflow and the local fan air supply remain unchanged in the next minute, the gas concentration in the next minute is calculated second by second, and the average value is taken as the prediction result of the theoretical model.
[0012] Furthermore, in step S3, the time series set of input data consists of gas concentrations at multiple historical moments. ,temperature T ( t ), air volume absolute atmospheric pressure P ( t and instantaneous productivity M ( t The input data is normalized and then input into the deep learning model.
[0013] Furthermore, in S4, the dynamic adjustment of weights is based on the root mean square error (RMSE) over the past 10 minutes, and the theoretical model is calculated accordingly. RMSE p and deep learning models RMSE d ; when RMSE p > RMSE d At the same time, increase the weights of the deep learning model. W d Reduce the weight of the theoretical model W p ; when RMSE p < RMSE d At that time, increase the weight of the theoretical model. W p Reduce the weights in deep learning models W d ; The W p + W d =1; The fusion formula is:
[0014] in, This is the final predicted gas concentration value. The prediction results of the theoretical model, This represents the prediction results from the deep learning model.
[0015] Furthermore, the formula for calculating the root mean square error is as follows:
[0016] in, N This represents the number of sampling points in the past 10 minutes. For the first i A predicted gas concentration value, For the first i The actual measured gas concentration value.
[0017] Furthermore, in step S5, the formula for calculating the prediction error is:
[0018] in, This refers to the actual monitored gas concentration value; When the prediction error exceeds a preset threshold, an optimization algorithm is used to adjust the weights and biases of the deep learning model.
[0019] Furthermore, the optimization algorithm is either the Adam optimization algorithm or the SGD optimization algorithm, which adjusts the parameters by calculating the gradient and updating them. The update formula is as follows:
[0020] in, For model parameters The new value, For model parameters New values, model parameters Including weights and biases, For learning rate, This is the gradient of the loss function.
[0021] Furthermore, the data is collected via a methane sensor, temperature sensor, air pressure sensor, air volume sensor, and a ground-based centralized control platform, with a collection frequency of once per second. The weights are recalculated and fused every minute, while an adaptive feedback mechanism is executed to continuously optimize the deep learning model.
[0022] The beneficial effects of this invention are as follows: (1) By organically integrating the theoretical model based on the gas diffusion equation with the deep learning model combining convolutional neural network and long short-term memory network, the physical mechanism's conservation constraints and interpretability on gas concentration changes are preserved, while the powerful ability of deep learning to capture nonlinear disturbances and long-short-term dependencies is fully utilized, which significantly improves the accuracy and reliability of gas concentration prediction in coal mine tunneling faces.
[0023] (2) A dynamic weighted fusion mechanism based on historical prediction error is introduced, which can adaptively adjust the weights according to the actual performance of the theoretical model and the deep learning model under different working conditions, so that the final prediction result always tends to the model with better performance. This gives full play to the advantages of the theoretical model when the production state is stable, and highlights the correction role of the deep learning model when there is a sudden gas outburst, thereby further improving the prediction accuracy.
[0024] (3) Set up an adaptive feedback mechanism, calculate the error and dynamically optimize the parameters of the deep learning model by comparing the predicted value with the actual monitoring value in real time, so that the model can quickly adapt to changes in the mine environment and differences in production disturbances, enhance the real-time performance and long-term stability of the prediction, and reduce the risk of prediction failure caused by model drift.
[0025] (4) Using multi-source data input, including parameters such as methane concentration, temperature, absolute atmospheric pressure, air volume and instantaneous productivity, the influencing factors of gas concentration are analyzed from multiple dimensions, avoiding the limitations of a single data source and improving the robustness and applicability of the prediction method.
[0026] (5) It has achieved accurate short-term prediction of gas concentration within the next minute, which has bought valuable response time for early warning of coal mine gas disasters, effectively reduced the risk of gas exceeding the limit and explosion, and provided more reliable technical support for mine safety production and miners' life safety.
[0027] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0028] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart of a method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning, according to the present invention. Detailed Implementation
[0029] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.
[0030] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.
[0031] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.
[0032] Figure 1 The system comprises the following modules: a multi-source data acquisition module for real-time acquisition of methane concentration, temperature, absolute atmospheric pressure, air volume, and instantaneous productivity; a theoretical model prediction module for preliminary prediction based on the gas diffusion equation; a deep learning model prediction module for nonlinear correction prediction using a combination of convolutional neural networks and long short-term memory networks; a dynamic weighted fusion module for adjusting weights based on historical errors and fusing the prediction results of the two models; an adaptive feedback module for comparing predicted and actual values and optimizing deep learning model parameters; and a final gas concentration prediction output module.
[0033] Example 1 This embodiment uses the 1203 fully mechanized tunneling face of a high-gas mine as the application object. The cross-sectional area of the roadway in this working face is 15m². 2 The monitored volume is taken within an 8m radius behind the tunnel head, i.e. The sensor network collects data in real time at a frequency of 1 Hz and transmits it to the ground data processing center via an industrial ring network.
[0034] exist At that moment, the real-time parameter collected was: total gas emission. Local fan air supply volume Current gas concentration The temperature is 26.5℃, the absolute atmospheric pressure is 101.2kPa, and the instantaneous productivity corresponds to a full-load cutting state of about 2.5t / min.
[0035] The workflow is as follows: First, the above parameters, including methane concentration, temperature, absolute atmospheric pressure, air volume, and instantaneous productivity, are collected through a multi-source data acquisition module.
[0036] Secondly, the theoretical model prediction module uses the gas diffusion equation for preliminary prediction. The equation is: (1) in, for The gas concentration at any given time, for Total gas emission at any given moment for The air volume supplied by the local fan at any given time. The volume of the tunneling face is monitored.
[0037] Substitute the collected parameters into equation (2) to calculate the concentration change rate: (2) (3) Note: A negative result indicates that the concentration decreases under ideal physical conditions.
[0038] Assuming in the next minute and The gas concentration remains unchanged. The iterative formula for calculating the gas concentration within the next minute is: (The formula is repeated twice in the original text.) (4) in, The theoretical model predicts that the gas concentration after 1 minute will be approximately: (5) Through complete second-by-second iterative calculations, the predicted average methane concentration for the next minute by the theoretical model was obtained. .
[0039] Next, the deep learning model prediction module inputs the normalized time-series data from the past 10 minutes into a combined Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model. The CNN extracts local features; the computation process is as follows: (6) in, The convolution kernel weight matrix is... For bias, This generates local features. LSTM captures long-short-term dependencies, and the computation process is as follows: (7) in, This is a predicted gas concentration sequence for the next 1 minute. The model identifies instantaneous step increases in productivity and predicts a lagged surge in gas outbursts, outputting the average gas concentration for the next 1 minute. .
[0040] Then, the dynamic weighted fusion module backtracks 600 sampling points from the past 10 minutes and calculates the root mean square error: (8) get , .because Adjust weights , .
[0041] The fusion formula is: (9) Substitute the numerical values into the calculation: (10) Finally, the adaptive feedback module obtains the actual gas concentration after 1 minute. Calculation error The gradient is calculated and the parameters are updated using the Adam optimization algorithm: (11) in, For model parameters, The learning rate. The final gas concentration prediction output module is released. .
[0042] Example 2 This embodiment uses the 1105 longwall tunneling face of another mine as the object, with a roadway cross-sectional area of 18m². 2 Monitoring volume Production is stable.
[0043] Data collection parameters: , , .
[0044] The theoretical model prediction module uses equations (1) to (4) to calculate and obtain the results. .
[0045] Deep learning model prediction module output .
[0046] Dynamic weighted fusion module calculation , ,Adjustment , Using equation (9) we obtain .
[0047] actual concentration The adaptive feedback module has been slightly adjusted.
[0048] Example 3 This example is for a sudden air outage scenario. The air volume is 480m³ / h. 3 / min decreased to 200m 3 / min.
[0049] The theoretical model prediction module calculates using formula (1) .
[0050] Deep learning model prediction module output .
[0051] Dynamic weighted fusion module allocation , Using formula (9) to obtain .
[0052] The actual peak value was 0.67%, and the adaptive feedback module enhanced the model response.
[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning, characterized in that: Includes the following steps: S1: Collect multi-source data on methane concentration, temperature, absolute atmospheric pressure, air volume, and instantaneous productivity at the coal mine tunneling face. The data is transmitted to the data processing center in real time via a sensor network. S2: Based on the gas diffusion equation, a theoretical model is constructed, and the gas concentration change rate is calculated using the temperature, absolute atmospheric pressure, air volume and gas emission rate, so as to preliminarily predict the gas concentration in the next minute. S3: The collected real-time multi-source data is input into a deep learning model that combines a convolutional neural network (CNN) and a long short-term memory network (LSTM). The CNN extracts local features from the multi-source data, and the LSTM captures long-term and short-term dependencies, thereby predicting the gas concentration within the next minute. S4: Dynamically adjust the weights of the theoretical model and the deep learning model based on historical prediction errors, and perform weighted fusion of the prediction results of the theoretical model and the prediction results of the deep learning model to obtain the final gas concentration prediction value. S5: Through an adaptive feedback mechanism, the final predicted gas concentration is compared with the actual monitored gas concentration, the prediction error is calculated, and the parameters of the deep learning model are dynamically adjusted according to the prediction error.
2. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 1, characterized in that: In S2, the gas diffusion equation is: in, for t The gas concentration at any given time, for t Total gas emission at any given moment for t The air volume supplied by the local fan at any given time. V The volume of the tunneling face is monitored.
3. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 2, characterized in that: In S2, starting from the gas concentration at the current moment, assuming that the total gas outflow and local fan air supply remain unchanged in the next minute, the gas concentration in the next minute is calculated second by second, and the average value is taken as the prediction result of the theoretical model.
4. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 1, characterized in that: In step S3, the time series set of input data consists of gas concentrations at multiple historical moments. ,temperature T ( t ), air volume absolute atmospheric pressure P ( t and instantaneous productivity M ( t The input data is normalized and then input into the deep learning model.
5. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 1, characterized in that: In step S4, the dynamic weight adjustment is based on the root mean square error (RMSE) over the past 10 minutes, and the theoretical model is calculated accordingly. RMSE p and deep learning models RMSE d ; when RMSE p > RMSE d At the same time, increase the weights of the deep learning model. W d Reduce the weight of the theoretical model W p ; when RMSE p < RMSE d At that time, increase the weight of the theoretical model. W p Reduce the weights in deep learning models W d ; The W p + W d =1; The fusion formula is: in, This is the final predicted gas concentration value. The prediction results of the theoretical model, This represents the prediction results from the deep learning model.
6. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 5, characterized in that: The formula for calculating the root mean square error is: in, N This represents the number of sampling points in the past 10 minutes. For the first i A predicted gas concentration value, For the first i The actual measured gas concentration value.
7. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 1, characterized in that: In step S5, the formula for calculating the prediction error is: in, This refers to the actual monitored gas concentration value; When the prediction error exceeds a preset threshold, an optimization algorithm is used to adjust the weights and biases of the deep learning model.
8. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 7, characterized in that: The optimization algorithm is either the Adam optimization algorithm or the SGD optimization algorithm. Adjustment is achieved by calculating the gradient and updating the parameters. The update formula is as follows: in, For model parameters The new value, For model parameters New values, model parameters Including weights and biases, For learning rate, This is the gradient of the loss function.
9. The method for predicting gas concentration in coal mine tunneling faces based on a combination of theoretical models and deep learning as described in claim 1, characterized in that: The data is collected via a methane sensor, temperature sensor, air pressure sensor, air volume sensor, and ground-based centralized control platform, with a collection frequency of once per second. The weights are recalculated and fused every minute, while an adaptive feedback mechanism is executed to continuously optimize the deep learning model.