Automatic inclination adjusting method for working face planning of upper computer based on artificial coal mining process experience data model

Abstract

The application discloses an automatic inclination adjusting method for working face planning of an upper computer based on an artificial coal mining process experience data model, and is applied to the fields of intelligent coal mining and artificial intelligence control, and aims at the collaborative posture control problem of multiple large devices under dynamic geological conditions in the special environment of coal mines. First, geological data are collected in real time, including coal seam inclination, top and bottom plate shape, device displacement state and multiple source data; then, inclination adjustment decision data of experienced coal mining technical personnel under different geological conditions are collected to construct a deep neural network mapping model; then, based on real-time geological data and the experience mapping model, an inclination adjustment process planning diagram is automatically generated; further, inclination adjustment planning instructions are sent to a hydraulic support electro-hydraulic control system and a coal mining machine control system, process collaboration is implemented, and based on the inclination adjustment data, the experience mapping model is incrementally trained and parameter optimized through a reinforcement learning mechanism.

Description

Technical Field

[0001] This invention belongs to the field of intelligent coal mining and artificial intelligence control, and specifically relates to an automatic tilt adjustment technology. Background Technology

[0002] Incline adjustment of the fully mechanized mining face is a core control technology to prevent the scraper conveyor from slipping or sliding and to prevent hydraulic support crushing accidents. Traditional inclination adjustment operations mainly rely on on-site technicians to manually judge and adjust based on geological parameters such as equipment displacement and coal seam dip angle, which has the following prominent problems:

[0003] First, decision-making experience is difficult to quantify and pass on. The experience of senior coal mining technicians in making decisions on slope adjustment under different geological conditions is mostly in the form of "tacit knowledge," lacking systematic data collection and modeling methods, which leads to long training cycles for new employees and large fluctuations in the quality of slope adjustment.

[0004] Secondly, the adjustment of the tilting parameters has a significant lag. Manual tilting is usually initiated only after the displacement of the scraper conveyor exceeds the safety threshold. This response delay leads to increased abnormal wear of the equipment and, in severe cases, can cause safety accidents such as support frame extrusion and conveyor jamming.

[0005] Third, existing automatic tilt adjustment technologies lack adaptability. Currently, automatic tilt adjustment systems based on laser alignment sensors and support posture detection mostly adopt fixed threshold control strategies, lacking digital modeling and application of experience in manual tilt adjustment processes under complex geological conditions, making it difficult to achieve adaptive optimization of tilt adjustment parameters.

[0006] Fourth, the planning and slant adjustment control of the host computer are disconnected. Although the existing automated coal mining methods based on host computer planning have been applied to three-dimensional detection of the working face and precise positioning of the coal mining machine, the slant adjustment planning and the adjustment of cutting process parameters lack a coordinated optimization mechanism, resulting in poor coal wall flatness and large fluctuations in equipment load after slant adjustment.

[0007] Therefore, there is an urgent need to develop a host computer-based automatic slope adjustment method for working faces that integrates empirical data models of manual coal mining processes, so as to achieve intelligent, adaptive, and process-coordinated slope adjustment decisions. Summary of the Invention

[0008] To address the aforementioned technical problems, this invention discloses a host computer-based method for automatically adjusting the inclination of a working face based on a data model of manual coal mining process experience. By combining manual coal mining experience data, a "geology-equipment-process" feature mapping model is constructed to achieve intelligent planning and execution of inclination adjustment decisions. The model also has the ability to learn and evolve online, significantly improving the adaptability and reliability of automatic inclination adjustment of working faces under complex geological conditions.

[0009] The technical solution adopted in this invention is: a method for automatic inclination adjustment of working faces based on a host computer-aided planning system using empirical data models of manual coal mining processes, comprising:

[0010] S1. Collect geological parameters and equipment status, and record the corresponding tilt adjustment decisions;

[0011] S2. Construct feature vectors based on the geological parameters and equipment status collected in step S1, and construct label vectors based on the tilt adjustment decisions recorded in step S1.

[0012] S3. Construct a deep neural network mapping model; train the deep neural network mapping model by taking the feature vector as input and the label vector as output.

[0013] S4. Deploy and apply the trained deep neural network mapping model;

[0014] S5. Take the geological parameters, equipment status and corresponding tilt adjustment decisions that meet the tilt adjustment quality index generated during the deployment and application process as incremental samples, return to step S3, and perform incremental training on the deep neural network mapping model.

[0015] Step S4 includes:

[0016] S41. Determine whether skew adjustment is triggered based on the cumulative displacement of the scraper conveyor. If skew adjustment is triggered, proceed to step S42.

[0017] S42. Input the real-time collected geological parameters and equipment status into the trained deep neural network mapping model to obtain the corresponding tilt adjustment decision;

[0018] S43. Based on the length of the tilt adjustment segment in the current tilt adjustment decision. Calculate the starting position of the tilt section;

[0019] S44. Based on the recommended angle in the current tilt adjustment decision. And the starting position of the tilt segment obtained in step S43, to generate the target pseudo-tilt sequence;

[0020] S45. Based on the recommended angle in the current tilt adjustment decision. With the length of the tilt section Calculate the single-blade tilt adjustment amount;

[0021] S46. Based on the recommended angle in the current tilt adjustment decision. Calculate the compensation amount for the height of the coal mining machine drum;

[0022] S47. Based on the length of the tilt adjustment segment in the current tilt adjustment decision. The hydraulic supports are planned to be moved in groups and gradually.

[0023] S48. Based on the results obtained in steps S43-S47 and the difference in the number of cuts at the beginning and end of the tilt adjustment segment in the current tilt adjustment decision. Coal cutting path pattern identifier The coal mining machine and hydraulic support are coordinated and controlled. The displacement change of the scraper conveyor and the change of the pseudo-inclination angle before and after the inclination adjustment are calculated to obtain the inclination adjustment quality index. If the index is less than the threshold, the secondary inclination adjustment is triggered and step S49 is executed.

[0024] S49. Return to step S42, and use the real-time geological parameters and equipment status collected after the tilt adjustment as new feature vectors to input into the trained deep neural network mapping model to obtain the corresponding tilt adjustment decision.

[0025] The beneficial effects of this invention are as follows: First, this invention collects geological data in real time, including multi-source data such as coal seam dip angle, roof and floor morphology, and equipment displacement status. Then, it collects adjustment decision data from experienced coal mining technicians under different geological conditions and constructs a deep neural network mapping model. Next, based on the real-time geological data and the empirical mapping model, it automatically generates an adjustment process planning diagram. Finally, it sends the adjustment planning instructions to the hydraulic support electro-hydraulic control system and the coal mining machine control system to achieve coordinated process execution. Based on the adjustment data, it incrementally trains and optimizes the empirical mapping model through a reinforcement learning mechanism. This invention possesses significant technical advantages and application effects over existing technologies in the field of automatic adjustment technology for coal mine working faces, specifically reflected in the following aspects:

[0026] 1. This invention, for the first time, transforms the decision-making experience of senior coal mining technicians under different geological conditions into a computable and iterative deep neural network mapping model, solving the problem of automated systems "not knowing how to adjust" or "not daring to adjust" under complex working conditions. By constructing a three-dimensional feature vector of "geology-equipment-process", it realizes the quantitative expression and multi-dimensional fusion of adjustment decision factors, and significantly improves the accuracy of adjustment parameter recommendation compared with traditional control methods based on fixed thresholds.

[0027] 2. This invention breaks through the technical barrier of independent planning and skew control in the host computer system, and proposes a skew-truncation cooperative control algorithm. The model output... (Number of cuts) directly drives the adjustment of the coal mining machine's cutting frequency, combined with... (Coal cutting mode) The dynamic selection of cutting path strategy realizes the dual optimization of coal wall flatness and equipment load during the inclination process, which can significantly reduce the coal wall flatness error and equipment failure rate after inclination.

[0028] 3. This invention designs a dual triggering mechanism based on the cumulative displacement and speed of the scraper conveyor, enabling precise judgment and early intervention in skew adjustment. Compared to traditional "remedial" skew adjustment, this invention can trigger skew adjustment planning as soon as the displacement reaches the displacement threshold, avoiding abnormal equipment wear. Simultaneously, an S-shaped smooth transition curve is used to generate the target pseudo-skew angle sequence, ensuring the smoothness and continuity of the skew adjustment process.

[0029] 4. This invention proposes an evaluation index and grading method for tilt adjustment quality, enabling quantitative evaluation and visual display of tilt adjustment effects. Through a secondary tilt adjustment triggering mechanism, a closed-loop control process of "planning-execution-evaluation-optimization" is formed to ensure that the tilt adjustment quality always meets the requirements for safe production.

[0030] 5. This invention introduces an incremental training mechanism based on elastic weight consolidation, enabling online learning and continuous evolution of the empirical data model. Through dynamic updating of sample weights and model confidence evaluation, the system can absorb new tilting experience under new working conditions while retaining its adaptability to historical working conditions.

[0031] 6. This invention adopts a modular design architecture, supporting rapid adaptation to fully mechanized mining faces with different coal seam dip angles, working face lengths, and geological conditions. The system can be quickly deployed and applied simply by adjusting the inclination adjustment parameter range in the standard operation template library, demonstrating significant engineering promotion value.

[0032] 7. This invention achieves precise decoupled control of tilt adjustment process parameters. The abstract tilt adjustment target is decomposed into specific angles (…). ),scope( ) frequency ( ) and patterns ( These four dimensions enable the automated system to flexibly choose between "multi-blade tilt adjustment" or "large step tilt adjustment" strategies, just like a senior technician, significantly improving the success rate of tilt adjustment under complex geological conditions. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the modules of the system of the present invention;

[0034] Figure 2 This is a flowchart of the method of the present invention;

[0035] Figure 3 This is the timing diagram for the tilt-cutting coordinated control of the present invention;

[0036] Figure 4 This is a schematic diagram of the online evolution mechanism of the model of the present invention. Detailed Implementation

[0037] To facilitate understanding of the technical content of this invention by those skilled in the art, the following description, in conjunction with the accompanying drawings, further illustrates the invention.

[0038] like Figure 1 As shown, the system of the present invention includes the following six modules:

[0039] A. Geological Data Acquisition Module: Through equipment such as the working face 3D laser scanning system, inertial navigation system, and support attitude sensor, it collects multi-source data in real time, including coal seam dip angle, roof and floor morphology, and equipment displacement status.

[0040] B. Human Experience Data Modeling Module: Collects adjustment decision data from senior coal mining technicians under different geological conditions and constructs a deep neural network mapping model.

[0041] C. Upper computer tilting planning module: Based on real-time geological parameters, equipment status and the constructed experience mapping model, it automatically generates tilting process planning diagram.

[0042] D. Incline Adjustment-Cutting Coordinated Control Module: This module sends the inclination adjustment planning command to the hydraulic support electro-hydraulic control system and the coal mining machine control system to achieve coordinated execution of the process.

[0043] E. Incline adjustment effect evaluation module: Based on the displacement convergence of the scraper conveyor, calculate the inclination adjustment quality index and feed it back to the planning module.

[0044] F. Online Model Evolution Module: Records skew adjustment execution data and performs incremental training and parameter optimization on the empirical mapping model through a reinforcement learning mechanism.

[0045] like Figure 2 As shown, the method of the present invention includes the following steps:

[0046] 1. Collect geological parameters and equipment status, and record the corresponding tilt adjustment decisions;

[0047] The collected geological parameters include the true dip angle of the coal seam. Current pseudo-oblique angle of the working face Rate of change of height difference between top and bottom plates The collected equipment status data includes the displacement speed of the scraper conveyor. Average cutting resistance of coal mining machine Standard deviation of bracket tilt angle The recorded adjustment decision content includes: recommended angle. Length of the skew section Number of cuts added or subtracted Coal cutting mode identifier .

[0048] 2. Construct feature vectors based on the geological parameters and equipment status collected in step 1, and construct label vectors based on the tilt adjustment decisions recorded in step 1;

[0049] 21. Based on multi-source data such as coal seam dip angle, roof and floor morphology, and equipment displacement status collected by the geological data acquisition module, a feature vector is constructed. The method of this invention aims to establish a mapping relationship between "geology-equipment-process," first defining the feature vector. :

[0050]

[0051] in: This represents the true dip angle of the coal seam. The current pseudo-hedge angle of the working face. The percentage change in elevation difference between the top and bottom plates. The displacement speed of the scraper conveyor. This represents the average cutting resistance of the coal mining machine. The standard deviation of the bracket tilt angle is given by the superscript T, which indicates transposition.

[0052] 22. Based on the data collected by the human experience data modeling module, the decision-making data on slope adjustment by senior coal mining technicians under different geological conditions are used to construct a label vector; the label vector for slope adjustment process parameters is also constructed. Defined as:

[0053]

[0054] Among them: Recommendation angle The target change amount that determines the pseudo-angle of the working face;

[0055] Adjusting the length of the inclined section Determine the scope of hydraulic supports involved in tilt adjustment;

[0056] The difference in the number of cuts at the beginning and end of the tilting section (net number of cuts added or subtracted) is the value of the number of cuts on the head side of the machine and the number of cuts on the tail side of the machine. It directly determines the tilting rate.

[0057] Coal cutting path mode identifier A value of 1 indicates "unidirectional oblique cutting" (adjusting the advance amount only when the tool is advancing in one direction), a value of 2 indicates "bidirectional collaborative cutting" (adjusting the advance amount in both forward and reverse directions), and a value of 3 indicates "end-point enhanced cutting" (executing a special cutting path only in the end-point area).

[0058] 3. Construct a deep neural network mapping model:

[0059] This invention employs a multilayer perceptron (MLP) network. This achieves a nonlinear mapping from feature vectors to skewing parameters.

[0060]

[0061] in, Let L be the set of network parameters, and L be the number of network layers.

[0062] The multilayer perceptron network structure used in this embodiment includes:

[0063] Input layer: 6 neurons, corresponding to the 6 dimensions of the feature vector X;

[0064] Hidden layer 1: 24 neurons, activation function is ReLU:

[0065]

[0066] Hidden layer 2: 16 neurons, activation function is ReLU.

[0067]

[0068] Output layer: 4 neurons, activation function selected based on output type:

[0069]

[0070] Loss function: Weighted mean squared error (WMSE) is used as the training loss function.

[0071]

[0072] in, This represents the number of training samples; The true value of the j-th class skew label for the i-th sample (from human experience records); This is the model's predicted value for the j-th class skew label of the i-th sample (i.e., the output of the j-th dimension of the output layer). Corresponding to recommendation angles Length of the slant section Add or subtract the number of cuts Coal cutting mode identifier ; The weighting coefficients for the j-th type of tilting parameter are... The inherent engineering importance of the tilting parameter (i.e., it does not change over time) is considered, and the parameter is set based on its impact on the tilting quality. The training loss function here... The sample-level scalar coefficients are omitted. Because considering the weight of the same type of label across all samples Since the values ​​are consistent, only j is retained as the difference in the output dimension; that is, the training loss function retains j. .

[0073] Based on the feature vectors and label vectors from step 2, construct the training dataset:

[0074] The training data comes from intelligent working faces with complete historical data records and abundant tilt adjustment cases in several typical mining areas in recent years. Valid samples with successful tilt adjustment and meeting quality standards were selected. The specific selection principles are as follows:

[0075] Data integrity requirements: The geological data acquisition system, equipment status monitoring system, and tilt adjustment decision recording system of the selected working face must be running synchronously, with timestamp alignment accuracy ≤10s, to ensure clear causal relationships between features and labels;

[0076] Working conditions and coverage requirements: typical working conditions such as coal seam dip angle of 8°~35°, working face length of 150~320m, and roof stability of Class I~IV;

[0077] Sample validity requirements: Only successful cases with convergence of scraper conveyor displacement and a tilting quality index Q≥0.8 after tilting are selected to avoid noise samples interfering with model training.

[0078] (3) Data preprocessing

[0079] Feature normalization: Robust scaling is applied to continuous features in the training data to suppress the influence of outliers;

[0080] Time alignment: Geological data, equipment status, and tilt adjustment decision timestamps are aligned to a 10-second granularity to ensure clear causal relationships between features and labels.

[0081] Training configuration:

[0082] Dataset partitioning: Training set: Validation set: Test set = 7:2:1, stratified sampling according to working condition type to avoid data leakage;

[0083] Optimizer: AdamW, initial learning rate It employs a cosine annealing strategy, with a 50% decay rate every 20 rounds;

[0084] Early stopping mechanism: If the validation set loss does not decrease for 5 consecutive rounds, training is terminated and the optimal model is retained;

[0085] Model ensemble: Five sub-models with different initializations are trained, and a weighted average is taken during prediction to improve robustness.

[0086] The deep neural network mapping model was trained using the training dataset described above.

[0087] 4. Based on the output of the deep neural network mapping model , , , Conduct an accuracy assessment;

[0088] The "accuracy evaluation" aims to verify the engineering usability of the model on independent test sets. Its core logic is to calculate the model's prediction parameters. With real labels The model accuracy is considered satisfactory when the error is below the allowable threshold of the process. Given that the output dimension includes both continuous physical quantities and discrete process parameters, a differentiated index is used for comprehensive judgment. Accuracy evaluation requires the following conditions to be met simultaneously:

[0089] (1) Continuous quantity assessment: Recommended perspective The mean absolute error (MAE) is ≤1.5°; the length of the tilting section is... The average relative error RE ≤ 10%;

[0090] (2) Evaluation of discrete quantities: number of cuts added or subtracted The accurate matching rate (EMR) is ≥85%; coal cutting mode identification. The classification accuracy (ACC) is ≥90%.

[0091] (3) Overall convergence: The weighted mean square error of the dataset is WMSE ≤ 0.08.

[0092] The above threshold settings can be adjusted based on the experience tolerance of senior technicians on site to ensure that the model output has the engineering accuracy to be directly sent to the electro-hydraulic control system.

[0093] Taking a typical mining area's independent test set (N=200) as an example, the model evaluation results are as follows: The MAE is 1.2°. The RE was 7.8%. 174 cases were exact matches (EMR of 87.0%). The model correctly classified 186 cases (ACC of 93.0%), and the WMSE on the test set was 0.065. All four metrics were better than the set thresholds. Based on the overall assessment, the deep neural network mapping model passed the accuracy evaluation and can proceed to step 5, the deployment and application phase.

[0094] 5. Deploy and apply the deep neural network mapping model that has passed the accuracy evaluation. The application process is as follows:

[0095] 51. Inclination Triggering Conditions

[0096] Define the cumulative displacement of the scraper conveyor :

[0097]

[0098] in: The displacement increment during the kth sampling period.

[0099] Tilt Trigger Determination Function Defined as:

[0100]

[0101] in: Displacement threshold , For displacement velocity threshold, .

[0102] when If the value is 1, then proceed to step 52;

[0103] 52. Calculation of the start and end positions of the tilt section, and the starting frame number of the tilt section. With termination number Calculate using the following formula:

[0104]

[0105] in: The reference frame number for tilt adjustment is usually taken from the middle of the working face.

[0106] 53. Based on the output of the deep neural network mapping model and the result calculated in step 52 and Generate a target pseudo-oblique angle sequence and define the target pseudo-oblique angle of the i-th support within the oblique adjustment segment. :

[0107]

[0108] in: The pseudo-inclination angle of the working surface before inclination adjustment. This is a function for adjusting the slope curve;

[0109] An S-shaped smooth transition curve is used, expressed as follows:

[0110]

[0111] in: The midpoint of the slope adjustment section is denoted by k, which is the curve steepness coefficient, ranging from 0.2 to 0.5.

[0112] Meanwhile, the recommended angle is based on the output of the deep neural network mapping model. With the length of the tilt section Calculate the single-blade skew adjustment amount :

[0113]

[0114] If the deep neural network mapping model outputs The tilt can be adjusted by changing the step distance of the support; if the output of the deep neural network mapping model is... The angle adjustment is achieved by controlling the difference in cutting frequency at both ends of the inclined section of the coal mining machine.

[0115] 54. Coal mining machine drum height compensation: Recommended angle based on deep neural network mapping model output within the inclined section. Calculate the height compensation of the coal mining machine drum. :

[0116]

[0117] Where: K is an empirical coefficient, with a value ranging from 0.8 to 1.2. For location-related weighting functions, The expression is:

[0118]

[0119] 55. The hydraulic support is moved in groups and progressively, dividing the inclined section into groups G, each group containing... Frame support:

[0120]

[0121] The timing of the movement of the g-th group of stents Defined as:

[0122]

[0123] in: The time interval between shifts between groups is usually taken as 2 to 3 coal mining machine cutting cycles.

[0124] 56. The host computer inputs real-time geological parameters and equipment status into the trained deep neural network mapping model to obtain... (Difference in the number of additions and subtractions) and (Coal cutting path mode identifier). The aforementioned , Combined with the planning parameters calculated in the above steps, a collaborative control instruction set for the coal mining machine and hydraulic support is generated. The specific mapping relationship is as follows:

[0125] (1) Mapping of cutting frequency and geometric target: using the target pseudo-oblique angle sequence and single-blade oblique adjustment amount Due to spatial constraints, It is decomposed into differentiated cutting cycle commands for the head side and tail side of the coal mining machine. When At that time, the coal mining machine was operating on the head side. During the additional cutting stroke, the tail side maintains the baseline cutting; when The reverse additional cutting is performed in real time, and the pseudo-oblique angle of the working face is adaptively adjusted by using the difference in the advance speed at both ends.

[0126] (2) Coupling of path pattern and shift timing: Command and hydraulic support grouping and moving sequence Deep binding determines the synchronization logic between the truncation process and stent advancement. If (One-way tilt adjustment) The support group movement is triggered only during the cutting stroke in the same direction as the tilt adjustment; if (Two-way coordination) triggers push compensation proportionally for both round trips, ensuring consistent machine-frame movement rhythm.

[0127] (3) Dynamic compensation of drum height: During the execution of the above cutting command, the drum height compensation of the coal mining machine is adjusted. As a real-time feedforward quantity superimposed on the coal mining machine height adjustment control system, it ensures the accuracy of the roof and floor following the cutting angle.

[0128] Finally, the host computer sends the fused cutting command, height adjustment compensation command, and support pushing timing to the coal mining machine control system and the hydraulic support electro-hydraulic control system, respectively, to realize the multi-equipment collaborative execution of the inclination adjustment process.

[0129] 57. For example Figure 3 As shown, the timing process of the tilt-cutting coordinated control is as follows:

[0130] (1) Host computer planning stage: After the host computer issues the tilt adjustment start command, it calculates the number of tool additions and subtractions based on the output of the deep neural network mapping model. Coal cutting mode identifier Generate a cutting path plan, determine the number of cutting cycles and the support grouping and moving strategy within the inclined section;

[0131] (2) Coal mining machine cutting cycle stage: The coal mining machine control system receives the cutting command issued by the host computer and executes the cutting cycle within the inclined section. During each cutting process, the drum height compensation is dynamically calculated and adjusted. To ensure the accuracy of the top and bottom plates following the angle cut;

[0132] (3) Hydraulic support grouping and moving stage: The host computer issues moving instructions to each group of hydraulic supports in sequence according to the planned grouping strategy. The first group of supports at time... Upon receiving the shift command, the second set of supports was at time... Receive the shift command, and so on, the nth group of supports at time... Receive pushing instructions. Each group of supports executes pushing operations according to the timing requirements, in stages and segments, to achieve gradual tilt adjustment;

[0133] (4) Collaborative control closed loop: Through the time-series coordination of "supervisor planning → coal mining machine cutting → support grouping and pushing", the action rhythm of coal mining machine cutting and support pushing is matched during the tilt adjustment process, avoiding machine-support movement interference and ensuring the smooth execution of the tilt adjustment process.

[0134] 58. Adjustment quality indicators: Define the adjustment quality evaluation indicator Q:

[0135]

[0136] in: This represents the change in displacement of the scraper conveyor before and after tilting. The displacement value of the scraper conveyor before skew adjustment is measured by the stroke sensor. The displacement value of the scraper conveyor after skewing is measured by the stroke sensor; To allow the maximum displacement, ; This represents the change in the pseudo-skew angle of the scraper conveyor before and after skew adjustment. ; Maximum allowable angular deviation; weighting coefficient: , .

[0137] Grade determination of tilt adjustment effect:

[0138] 59. Secondary tilt adjustment triggering conditions: when the tilt adjustment quality index... ( When the target slope is within the quality threshold, the host computer determines that the first slope adjustment has not reached the convergence standard and triggers the second slope adjustment planning process. The system first collects real-time geological parameters and equipment status after the slope adjustment is executed, and inputs these as new feature vectors into the deployed deep neural network mapping model to obtain the corrected slope adjustment decision parameters. Then, based on the new parameters, it recalculates the target pseudo-angle sequence, single-blade slope adjustment amount, drum height compensation amount, and hydraulic support grouping and pushing sequence, and generates a new round of coal mining machine cutting instructions and support collaborative control instructions. This "status acquisition - model deduction - instruction generation - execution evaluation" process iterates cyclically until... Or the number of iterations reaches the preset limit. (Usually 2-3 times) to form a closed-loop tilt control logic, effectively suppressing the accumulation of residual tilt error and ensuring the smooth convergence of the scraper conveyor's posture.

[0139] 6. The implementation process of the online evolution module of the model is as follows:

[0140] A new sample set is constructed based on the adjusted coal seam dip angle, roof and floor morphology, equipment displacement status, and corresponding adjustment decision data that meet the adjustment quality indicators. This module achieves online incremental training and continuous evolution of the deep neural network mapping model through the following sub-steps:

[0141] 61. Assign dynamic weights to historical samples and incremental samples. This allows for dynamic adjustment of the contribution of each sample error term to the model parameter updates within the loss function. The dynamic weight expression is:

[0142]

[0143] in, The sample decay coefficient, with a value of 0.01 to 0.05, is used to control the rate of forgetting historical operating conditions and experience. The interval (in days) between sample i and the current time; This is the weight enhancement coefficient for new samples, ranging from 0.2 to 0.5, used to increase the learning priority of the current new data acquisition conditions; For indicator functions, when the sample belongs to the incremental dataset The value is 1 if the condition is right and 0 otherwise. This mechanism ensures that while the model absorbs the experience of adjusting the slope under new geological conditions, it weakens the interference of old conditions by decaying over time, thus achieving "smooth transition of historical experience - priority adaptation of new knowledge".

[0144] 62. The Elastic Weight Consolidation (EWC) method is used to construct the incremental training loss function to prevent catastrophic forgetting when the model learns new working condition data. The incremental training loss function is expressed as:

[0145]

[0146] in, For the k-th network parameter The importance coefficient is obtained by approximating the diagonal elements of the Fisher information matrix using historical samples. A larger value indicates that the parameter is more sensitive to historical tilting decisions, and its offset needs to be limited during incremental training. These are the old model parameters, which are the snapshots of the optimal parameters saved after the model has completed the previous stage of training and passed the accuracy evaluation. They represent the solidified experience under historical conditions. The regularization coefficient has a range of values. , used to balance the constraint strength between "learning new knowledge" and "retaining old experience"; k is the index of trainable parameters of the network.

[0147] In the incremental training loss function here Compared to The weighting coefficients reflecting the inherent engineering importance of the tilt adjustment parameters have been omitted. Preserve the time-dependent value of samples under the emphasis on the time-varying characteristics of geological conditions (dynamic decay), that is, retain dynamic weights. .

[0148] By training the loss function incrementally, highly important parameters will be adjusted during backpropagation. Applying stronger regularization penalties forces the model to "flexibly deform" rather than "hardly cover" historical experience when updating parameters, thereby maintaining the stability and continuity of tilt adjustment decisions under complex geological conditions.

[0149] 63. Model confidence evaluation, defining the model prediction confidence level. :

[0150]

[0151] in, For the validation set, This represents the Gaussian kernel bandwidth.

[0152] like Figure 4 As shown, when In this embodiment, or when a sudden change in geological parameters is detected, a manual intervention interface is opened. .

[0153] Those skilled in the art will recognize that the embodiments described herein are for the purpose of helping to understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Various modifications and variations can be made to the invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the scope of the claims of the invention.

Claims

1. A method for automatic inclination adjustment of working faces based on empirical data models of manual coal mining processes, characterized in that: include: S1. Collect geological parameters and equipment status, and record the corresponding tilt adjustment decisions; S2. Construct feature vectors based on the geological parameters and equipment status collected in step S1, and construct label vectors based on the tilt adjustment decisions recorded in step S1. S3. Construct a deep neural network mapping model; train the deep neural network mapping model by taking the feature vector as input and the label vector as output. S4. Deploy and apply the trained deep neural network mapping model; S5. Take the geological parameters, equipment status and corresponding tilt adjustment decisions that meet the tilt adjustment quality index generated during the deployment and application process as incremental samples, return to step S3, and perform incremental training on the deep neural network mapping model.

2. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 1, characterized in that, The geological parameters in step S1 include: the true dip angle of the coal seam. The current pseudo-oblique angle of the working face Rate of change of height difference between top and bottom plates The equipment status in step S1 includes: the displacement speed of the scraper conveyor. Average cutting resistance of coal mining machine Standard deviation of bracket tilt angle Step S1, the tilt adjustment decision includes: recommending the angle. Length of the skew section The difference in the number of cuts at the beginning and end of the inclined section Coal cutting path mode identifier .

3. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 2, characterized in that, The deep neural network mapping model in step S3 specifically uses a multilayer perceptron network to achieve a non-linear mapping from feature vectors to label vectors.

4. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 3, characterized in that, The training loss function used in step S3 is: ； in, This represents the number of training samples; The true value of the j-th class skew label for the i-th sample; This represents the predicted value of the j-th class tilt label for the i-th sample by the deep neural network mapping model. Each corresponds to a recommendation angle. Length of the skew section Add or subtract the number of cuts Coal cutting mode identifier ; is the weighting coefficient of the j-th type of tilting parameter.

5. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 4, characterized in that, Step S4 includes: S41. Determine whether skew adjustment is triggered based on the cumulative displacement of the scraper conveyor. If skew adjustment is triggered, proceed to step S42. S42. Input the real-time collected geological parameters and equipment status into the trained deep neural network mapping model to obtain the corresponding tilt adjustment decision; S43. Based on the length of the tilt adjustment segment in the current tilt adjustment decision. Calculate the starting position of the tilt section; S44. Based on the recommended angle in the current tilt adjustment decision. And the starting position of the tilt segment obtained in step S43, to generate the target pseudo-tilt sequence; S45. Based on the recommended angle in the current tilt adjustment decision. With the length of the tilt section Calculate the single-blade tilt adjustment amount; S46. Based on the recommended angle in the current tilt adjustment decision. Calculate the compensation amount for the height of the coal mining machine drum; S47. Based on the length of the tilt adjustment segment in the current tilt adjustment decision. The hydraulic supports are planned to be moved in groups and gradually. S48. Based on the results obtained in steps S43-S47 and the difference in the number of cuts at the beginning and end of the tilt adjustment segment in the current tilt adjustment decision. Coal cutting path pattern identifier The coal mining machine and hydraulic support are coordinated and controlled. The displacement change of the scraper conveyor and the change of the pseudo-inclination angle before and after the inclination adjustment are calculated to obtain the inclination adjustment quality index. If the index is less than the threshold, the secondary inclination adjustment is triggered and step S49 is executed. S49. Return to step S42, and use the real-time geological parameters and equipment status collected after the tilt adjustment as new feature vectors to input into the trained deep neural network mapping model to obtain the corresponding tilt adjustment decision.

6. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 5, characterized in that, In step S41, if the cumulative displacement of the scraper conveyor is greater than the set displacement threshold, or if the displacement speed calculated based on the cumulative displacement of the scraper conveyor is greater than the set displacement speed threshold, then skew adjustment is triggered.

7. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 6, characterized in that, Step S44 uses an S-shaped smooth transition curve to generate the target pseudo-oblique angle sequence.

8. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 7, characterized in that, The formula for calculating the tilt adjustment quality index in step S47 is: ； in: This represents the change in displacement of the scraper conveyor before and after tilting. To allow the maximum displacement, This represents the change in the pseudo-skew angle of the scraper conveyor before and after skew adjustment. , All are weighting coefficients.

9. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 8, characterized in that, The incremental training loss function used in step S5 is: ； in, The dynamic weights corresponding to the i-th sample are: For the k-th network parameter Importance coefficient, For the old model parameters, This is the regularization coefficient.

10. The method for automatic inclination adjustment of the working face based on the empirical data model of manual coal mining technology according to claim 9, characterized in that, The expression is: ； in, The sample attenuation coefficient; The interval between sample i and the current time; The weight enhancement coefficient for the new sample; This is an indicator function, meaning it takes the value 1 when the sample belongs to the incremental dataset, and 0 otherwise.

Images