Fully mechanized coal mining face roof pressure prediction method
By applying LightGBM model and hydraulic support sensor data on the comprehensive mining working face and dynamically adjusting the characteristics and models, the accuracy and real-time problems of underground mine pressure prediction are solved, efficient roof pressure prediction and early warning are achieved, and the effect of mine pressure management is improved.
Patent Information
- Application Number
- CN202510364623.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-26
- Publication Date
- 2025-07-18
AI Technical Summary
There are problems of fuzzy geological mechanism and dynamic coupling of multiple factors in the ore pressure prediction task of comprehensive mining working face. It is difficult for the existing technology to achieve accurate and timely pressure prediction, which affects the accuracy of ore pressure prediction and the effectiveness of the governance plan.
The LightGBM model is used to collect data in real time with hydraulic support pressure sensors, extract features through data preprocessing, establish training data sets, and automatically measure the importance of features during model training, dynamically adjust the model to adapt to changes in downhole conditions, and combine the support action and spatial position information to achieve pressure prediction and early warning.
It improves the accuracy and stability of mine pressure prediction, provides real-time pressure warning and disposal measures, and improves the decision-making basis and engineering efficiency of rooftop management.
Smart Images
Figure CN120337516A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of mine pressure, and specifically refers to a method for predicting the roof pressure of a fully mechanized coal mining face. Background Art
[0002] The accuracy of the mine pressure prediction results in a fully mechanized coal mining face is mainly affected by two factors: on the one hand, the underground environment is complex, and the pressure value is actually the result of the combined action of multiple factors. Many factors such as the lithology and thickness of the overlying strata, the position of the support, and the mining technology can directly affect the pressure borne by the hydraulic support; on the other hand, the working face strata are non-homogeneous and anisotropic, and as the mining progresses, all the factors affecting the mine pressure are changing dynamically. In this case, it is particularly important to grasp the key characteristics that determine the mine pressure and establish a mine pressure prediction scheme that can be continuously adjusted as the underground conditions change so as to grasp the pressure distribution of the working face in real time. In addition, accurate and timely pressure prediction results can also leave sufficient room for the implementation of treatment measures.
[0003] Underground pressure prediction technologies can generally be divided into three categories in principle, namely pure mathematical fitting methods, physical model-driven methods, and data-driven methods.
[0004] Mathematical fitting methods do not consider or consider less the scenarios of algorithm application, and only process the mine pressure sequence data through various sequence fitting means such as establishing various curve templates, various polynomial fittings, least squares fitting and its variants. These methods are effective but have limited effects, and can no longer meet the needs of the current various tasks for the accuracy of pressure prediction.
[0005] Physical model-driven methods introduce a real physical model into the process of pressure prediction. First, determine each parameter including formation parameters, and then use knowledge in related fields such as geology and mechanics to model the support pressure data. Since physical model-driven methods take into account the real physical process, the effect is better than traditional data fitting means. Also for this reason, the calculation complexity of the physical model-driven pressure prediction scheme is high and the efficiency is low. In addition, the formation information is complex and the geological conditions vary greatly, which pose challenges to physical modeling.
[0006] Most data-driven methods utilize the powerful mapping ability of neural networks, design a suitable network structure, and through the training of a large-scale data set, solidify the operator from each feature to the pressure value in the network weights to achieve the prediction of pressure. However, such methods face some problems. A large amount of time and energy need to be invested in constructing the data set and training the model in the early stage. The complexity of the above process leads to uneven effects of such methods; in addition, large models have poor interpretability, unclear principles, poor portability, and also face the generalization problem under different scenarios.
[0007] In summary, each processing method has its potential problems. How to combine the advantages of various solutions, make up for their disadvantages, and find a real-time pressure prediction solution suitable for the fully mechanized coal mining face is a difficult point. Summary of the Invention
[0008] (I) Technical Problem
[0009] The present invention aims to at least solve the problems of fuzzy geological mechanism and multi-factor dynamic coupling faced by the mine pressure prediction task under complex conditions in the fully mechanized coal mining face, improve the accuracy of pressure prediction, and provide a prerequisite for the good application of treatment schemes for strong pressure manifestation and other situations.
[0010] (II) Technical Content
[0011] This solution provides a roof pressure prediction system for a fully mechanized coal mining face, including a pressure acquisition module, a data preprocessing module, a model training module, and a pressure prediction module;
[0012] The pressure acquisition module collects the original roof pressure data in real time through high-precision pressure sensors built in hydraulic supports, with a high data acquisition frequency, and synchronously obtains the support action status and spatial position information;
[0013] The data preprocessing module is connected to the pressure acquisition module and is used to clean and transform the original pressure data; extract data features related to pressure value prediction and establish a training data set;
[0014] The model training module is used to establish a LightGBM model and train the model using the training data set;
[0015] The pressure prediction module feeds back the prediction result through the trained LightGBM model.
[0016] Preferred Technical Solution 1: It further includes a model update module, and the model update module is used to adjust and optimize the model.
[0017] Preferred Technical Solution 2: It further includes an early warning module. The prediction result of the pressure prediction module is fed back to the early warning module in real time, and at the same time, the early warning module is used to give an early warning and corresponding disposal measures according to the pressure prediction result.
[0018] Preferred Technical Solution 3: It further includes a database, and the data collected by the pressure acquisition module is stored in the local database in a time series form.
[0019] Preferred Technical Solution 4: It further includes a feature library, and the feature library is used to store features related to pressure value prediction.
[0020] This solution also discloses a roof pressure prediction method for a fully mechanized coal mining face, including the following steps:
[0021] (1) Step S1: Determine the data features associated with pressure value prediction
[0022] Step S11: Collect the pressure sequence data of the hydraulic support
[0023] A hydraulic support is a power device that uses hydraulic pressure to generate a supporting force for working face roof support and management. During the fully mechanized coal mining process, the pressure generated by the overlying strata on the working face acts on the hydraulic support, and the pressure sensor inside the hydraulic support collects the data of roof weighting and feeds it back to the pressure acquisition module;
[0024] The pressure data changes over time under the influence of factors such as the state of the overlying strata and the actions of the support, presenting as time series data with periodic characteristics;
[0025] Step S12: Preprocess the pressure sequence data of the hydraulic support
[0026] The pressure sequence data collected by the pressure acquisition module and stored in the database is rough. Before extracting features, it needs to be preprocessed using a data preprocessing module. The preprocessing process includes outlier removal, data resampling, data interpolation, data filtering, etc.;
[0027] Step S13: Extract the data features related to pressure value prediction and establish a training dataset
[0028] Select the preprocessed pressure sequence data, perform statistical analysis on it to obtain the features related to pressure value prediction, and combine other features to form a feature set; use the feature set as the input and the set of pressure values to be predicted as the output to establish a training dataset;
[0029] Select features to enter the feature set according to different scenarios and conditions;
[0030] (2) Step S2: Establish a mine pressure prediction model
[0031] Step S21: Create a LightGBM model
[0032] Create a LightGBM model, and the model training module uses the mean squared error as the loss function to measure the training effect;
[0033] Step S22: Train the LightGBM model
[0034] During the training process, the importance of features can be automatically measured to determine which features have a significant impact on the accurate prediction of the model; if a certain feature has little impact on pressure prediction, return to step S13 to remove the feature from the feature set and re - execute the subsequent process;
[0035] Step S23: A pressure prediction system was established with the LightGBM model as the core;
[0036] (3) Step S3: Optimize the LightGBM model during the operation of the pressure prediction system
[0037] During the mining process, various factors affecting the mine pressure are changing dynamically, and the weight of each factor will also change. Therefore, the elements in the feature set should not be static; Select the maintenance shift time or other appropriate time periods that do not affect production to re-execute Steps S1 and S2 to achieve dynamic adjustment of the model;
[0038] (4) Step S4: Give early warnings and corresponding disposal measures according to the pressure prediction results
[0039] Step S41: Give mine pressure warning information
[0040] According to the pressure prediction results, using different pressure thresholds as trigger conditions, give corresponding-level pressure warning information at the front end of the system;
[0041] Step S42: Give corresponding disposal measures
[0042] According to the pressure prediction results and the pressure warning information, retrieve the corresponding disposal measures from the database.
[0043] (III) Technical effects
[0044] Adopting the above structure enables this solution to have the following beneficial effects:
[0045] 1. Applying the LightGBM algorithm to the mine pressure prediction task in the fully mechanized coal mining face effectively alleviates the negative impact of the mutual coupling of multiple factors underground on the pressure prediction accuracy, providing a decision-making basis for the roof control and management of the working face;
[0046] 2. Introduce the support action information and the support spatial position information into the pressure prediction process, establish a complete feature database, and select high-weight features as inputs to train the LightGBM model;
[0047] 3. Adopt the strategy of model dynamic optimization, update the model within a specific time period, improve the stability and generalization performance of the entire solution, and enable the model accuracy to remain stable in different scenarios. Description of the drawings
[0048] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:
[0049] Figure 1 is the system architecture diagram of this solution;
[0050] Figure 2 is the process schematic diagram of this solution;
[0051] Figure 3 is the importance measure of different features on the prediction result of mine pressure under a specific embodiment of this solution;
[0052] Figure 4 is the comparison between the predicted value and the actual value of mine pressure under a specific embodiment of this solution.
[0053] Among them, 1. Pressure acquisition module, 2. Data preprocessing module, 3. Model training module, 4. Pressure prediction module, 5. Model update module, 6. Early warning module. Specific implementation manner
[0054] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
[0055] Please refer to Figure 1 , the roof pressure prediction method for fully mechanized mining face, including a pressure acquisition module 1, a data preprocessing module 2, a model training module 3, and a pressure prediction module 4;
[0056] The pressure acquisition module 1 collects the original roof pressure data in real time through the high-precision pressure sensors built in the hydraulic supports, with a high data acquisition frequency, and synchronously obtains the support action states (such as information on raising and lowering the supports) and spatial position information (support number, working face coordinates);
[0057] The data preprocessing module 2 is connected to the pressure acquisition module 1 and is used to clean and transform the original pressure data; extract the data features related to the pressure value prediction and establish a training data set;
[0058] Specifically, the data preprocessing module 2 uses a sliding window combined with the Z-score algorithm to eliminate outliers, sets 60 MPa as the upper limit to filter sensor false alarms, and realizes data normalization through resampling and linear interpolation; extracts the pressure mean, variance, and trend within 5 - 30 minutes before time T, and constructs a 15-dimensional dynamic feature library in combination with the support action number, the pressure correlation of adjacent supports, etc.; the feature library can be flexibly adjusted and combined according to the model feedback (such as feature importance evaluation) to improve the generalization ability of the algorithm for different geological conditions;
[0059] The model training module 3 is used to establish a LightGBM model and train the model using the training data set;
[0060] Specifically, the model training module 3 is based on preprocessed data, and the mean square error (MSE) is used as the loss function to initialize the LightGBM model. The hyperparameters are optimized through cross-validation, and a single training takes only 172 seconds (i7-12700H CPU). During the training process, the feature importance (Gain value) is automatically calculated, and high-contribution features (such as support movement and spatial position) are screened, which significantly reduces the high training cost and low interpretability defects of neural network models, forming a lightweight and efficient core algorithm for mine pressure prediction.
[0061] The pressure prediction module 4 feeds back the prediction results through the trained LightGBM model;
[0062] Specifically, the pressure prediction module 4 uses the trained LightGBM model, inputs real-time feature data, outputs pressure prediction values at multiple future time points (such as T+1 to T+30 minutes), and supports spatial dimension prediction (such as changes in pressure as mining progresses) in combination with footage information; the prediction results are stored in the historical database for analysis and optimization, and its multi-step prediction capability provides a more ample response time window for emergency response;
[0063] It also includes a model updating module 5, which is used to adjust and optimize the model;
[0064] Specifically, in the scenario of dynamic changes in underground geological conditions, the model update module 5 triggers the incremental training process through maintenance shifts or low-load periods, iteratively optimizes model parameters and feature sets based on new data, and avoids repeated training of the full amount of data; combined with the latest feature importance evaluation results, it dynamically eliminates low-contribution features (such as redundant time statistics) to ensure the adaptability of the model to changes in mining technology and rock formations, and solves the problem of prediction failure caused by environmental changes in traditional static models;
[0065] It also includes an early warning module 6, and the prediction results of the pressure prediction module 4 are fed back to the early warning module 6 in real time. At the same time, the early warning module 6 is used to give early warnings and corresponding disposal measures according to the pressure prediction results;
[0066] Specifically, the pressure prediction module 4 sets multi-level thresholds (such as yellow warning 40-50MPa, red warning>50MPa) according to the predicted pressure value, displays the warning level and the affected support position in real time through a visual interface, and automatically retrieves the preset disposal plan (such as reinforced support, adjustment of mining speed) through the database, and pushes it to the on-site terminal; this mechanism forms a "prediction-response" closed loop, and combines feature importance analysis (such as SHAP value) to intuitively display key influencing factors, significantly improving the efficiency of roof accident prevention and control and the reliability of engineering decision-making;
[0067] It also includes a database. The data collected by the pressure acquisition module 1 is stored in the local database in a time series form, providing a multi-dimensional basis for subsequent feature extraction by combining multi-source data, avoiding the limitations of traditional methods that only rely on a single pressure sequence, and ensuring data integrity and scenario adaptability;
[0068] It also includes a feature library, which is used to store features related to pressure value prediction.
[0069] This solution also discloses a method for predicting the roof pressure of a fully mechanized coal mining face, including the following steps:
[0070] (1) Step S1: Determine the data features associated with pressure value prediction
[0071] Step S11: Collect the pressure sequence data of the hydraulic support
[0072] The hydraulic support is a power device that uses hydraulic pressure to generate a supporting force for roof support and management in the working face. During the fully mechanized coal mining process, the pressure generated by the overlying strata on the working face acts on the hydraulic support, and the pressure sensors inside the hydraulic support collect the data of roof weighting and feedback it to the pressure acquisition module 1;
[0073] The pressure data changes over time under the influence of factors such as the state of the overlying strata and the actions of the support, presenting as time series data with periodic characteristics;
[0074] Step S12: Preprocess the pressure sequence data of the hydraulic support
[0075] The pressure sequence data collected by the pressure acquisition module 1 and stored in the database is rough and needs to be preprocessed using the data preprocessing module 2 before feature extraction. The preprocessing process includes outlier removal, data resampling, data interpolation, data filtering, etc.;
[0076] In a specific embodiment, a. pressure values greater than 60 MPa; b. outlier pressure values are regarded as abnormal pressure values and removed from the pressure sequence;
[0077] For case a, in this embodiment, the pressure values greater than 60 MPa are due to sensor false alarms rather than normal high-pressure values;
[0078] For case b, a sequence window with a length of 100 is defined. The window is slid on the pressure sequence to select sample sequences. The Z-score is used to measure the standard deviation distance between each data point in the sample window and the mean of the sample window data, and the points with a Z-score greater than 3 are regarded as outliers and removed;
[0079] In this embodiment, the pressure data is resampled into a sequence with a time interval of 1 minute. During this process, data interpolation and filtering are achieved, and the interpolation method can choose average interpolation;
[0080] Step S13: Extract data features related to pressure value prediction and establish a training dataset
[0081] Select the preprocessed pressure sequence data, perform statistical analysis on it to obtain features related to pressure value prediction, and jointly form a feature set with other features; use the feature set as the input and the set of pressure values to be predicted as the output to establish a training dataset;
[0082] In a specific embodiment, define a prediction reference time T, and the feature set consists of the following features:
[0083] a. Respectively take the pressure sequences within the time periods of 5, 10, 20, and 30 minutes before time T, and calculate the mean, variance, and trend of the pressure sequences; use the mean of the data percentage change rate to measure the change trend of the pressure sequence; the above statistical process obtains 12 types of features;
[0084] b. The action number of the last action of the hydraulic support before time T; the actions of the support include lifting the column, lowering the column, moving the support, etc.;
[0085] c. The action number of the hydraulic support, which is used to measure the spatial difference of the hydraulic support in the working face;
[0086] d. The pressure value of the hydraulic support at time T;
[0087] The above process obtains a total of 15 types of features, which form a feature set. Use them as the input of the training set, and optionally use the pressure values at times T + 1, T + 5, T + 10, T + 20, and T + 30 as the output to respectively construct a training set;
[0088] The above feature set is for the pressure prediction task in the time dimension; if the pressure prediction task in the spatial dimension (predicting the change of pressure with the advance of the footage) is to be realized, the footage information should be added to the feature set; in addition, a relatively large feature library is established, and features are selected to enter the feature set according to different scenarios and conditions;
[0089] (2) Step S2: Establish a mine pressure prediction model
[0090] Step S21: Create a LightGBM model
[0091] Create a LightGBM model, and the model training module 3 uses the mean squared error as the loss function to measure the training effect;
[0092] Step S22: Train the LightGBM model
[0093] Training the model: In a specific embodiment, the training time on an i7-12700H CPU is 172s; during the training process, the feature importance can be automatically measured to determine which features have a significant impact on the accurate prediction of the model; if a certain feature has little impact on the pressure prediction, the process can return to step S13 to remove the feature from the feature set and re-execute the subsequent process;
[0094] Step S23: A pressure prediction system is established with the LightGBM model as the core;
[0095] (3) Step S3: Optimize the LightGBM model during the operation of the pressure prediction system
[0096] During the mining process, all factors affecting the mine pressure are changing dynamically, and the weight of each factor will also change. Therefore, the elements in the feature set should not be fixed; select the maintenance shift time or other appropriate time periods that do not affect production and re-execute steps S1 and S2 to achieve dynamic adjustment of the model;
[0097] (4) Step S4: Give early warnings and corresponding disposal measures according to the pressure prediction results
[0098] Step S41: Give mine pressure warning information
[0099] According to the pressure prediction results, using different pressure thresholds as trigger conditions, give pressure warning information of corresponding levels at the front end of the system;
[0100] Step S42: Give corresponding disposal measures
[0101] According to the pressure prediction results and pressure warning information, retrieve the corresponding disposal measures from the database.
[0102] The parts not disclosed in the present invention are all prior arts, and their specific structures and working principles will not be elaborated herein.
[0103] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A roof pressure prediction system for a fully mechanized mining face, characterized in that, It includes a pressure acquisition module (1), a data preprocessing module (2), a model training module (3), and a pressure prediction module (4); The pressure acquisition module (1) collects the original roof pressure data in real time through a high-precision pressure sensor built in the hydraulic support, with a high data acquisition frequency, and synchronously obtains the support action state and spatial position information; The data preprocessing module (2) is connected to the pressure acquisition module (1) and is used to clean and transform the original pressure data; extract data features related to the pressure value prediction and establish a training data set; The model training module (3) is used to establish a LightGBM model and train the model using the training data set; The pressure prediction module (4) feeds back the prediction result through the trained LightGBM model.
2. The fully-mechanized mining face roof pressure prediction system according to claim 1, wherein, It also includes a model update module (5), and the model update module (5) is used to adjust and optimize the model.
3. The roof pressure prediction system for fully mechanized coal mining face according to claim 1, wherein, It also includes an early warning module (6). The prediction result of the pressure prediction module (4) is fed back to the early warning module (6) in real time. At the same time, the early warning module (6) is used to give an early warning and corresponding disposal measures according to the pressure prediction result.
4. The fully-mechanized mining face roof pressure prediction system according to claim 1, characterized in that, It also includes a database, and the pressure sequence data collected by the pressure acquisition module (1) is stored in the local database in a time series form.
5. The roof pressure prediction system for fully-mechanized coal mining face according to claim 1, characterized in that It also includes a feature library, and the feature library is used to store features related to the pressure value prediction.
6. A roof pressure prediction method for a fully mechanized coal mining face according to claims 1-5, characterized in that, It includes the following steps: (1) Step S1: Determine the data features associated with the pressure value prediction Step S11: Collect the pressure sequence data of the hydraulic support The pressure sensor inside the hydraulic support collects the roof pressure sequence data and feeds it back to the pressure acquisition module (1); Step S12: Preprocess the pressure sequence data of the hydraulic support Before extracting features from the pressure sequence data collected by the pressure acquisition module (1) and stored in the database, it needs to be preprocessed using the data preprocessing module (2). The preprocessing process includes outlier removal, data resampling, data interpolation, and data filtering; Step S13: Extract data features related to the pressure value prediction and establish a training data set Select the preprocessed pressure sequence data, perform statistical analysis on it to obtain features related to the pressure value prediction, and jointly form a feature set with other features; Using the feature set as the input and the set of pressure values to be predicted as the output, establish a training data set; Select features to enter the feature set according to different scenarios and conditions; (2) Step S2: Establish a mine pressure prediction model Step S21: Create a LightGBM model Create a LightGBM model, and the model training module (3) uses the mean squared error as the loss function to measure the training effect; Step S22: Train the LightGBM model During the training process, measure the feature importance and remove features that have little impact on the pressure prediction from the feature set; Step S23: Establish a pressure prediction system with the LightGBM model as the core; (3) Step S3: Optimize the LightGBM model during the operation of the pressure prediction system; (4) Step S4: Give an early warning and corresponding disposal measures according to the pressure prediction result; Step S41: Give mine pressure early warning information According to the pressure prediction results, different pressure thresholds are used as triggering conditions to give pressure warning information of corresponding levels; Step S42: Give corresponding disposal measures According to the pressure prediction results and the pressure warning information, corresponding disposal measures are retrieved from the database.