Coal mine accurate spraying control method based on point cloud segmentation and dust prediction

Abstract

The invention belongs to the technical field of intelligent mines, particularly relates to a coal mine precise spraying control method based on point cloud segmentation and dust prediction, and aims to improve the intelligence and refinement level of coal mine dust treatment. Comprising the steps of collecting three-dimensional point cloud data of a coal mine operation environment in real time through a laser radar, performing semantic segmentation on the three-dimensional point cloud data by using a deep learning model, identifying a dust distribution area, mine airflow parameters and operation intensity information, and constructing a dynamic space scene semantic graph. And performing real-time prediction on a dust diffusion path, a concentration change trend and a high-risk area by adopting an AI prediction model fused with spatio-temporal characteristics based on the dynamic space scene semantic graph. And according to the dust prediction result, the on-site working condition and the spraying system layout are combined, a start-stop instruction, spraying direction parameters and strength parameters of the intelligent spraying device are generated, dust falling effect data are obtained, the dust falling effect data are transmitted back to the control end through a closed-loop feedback mechanism, and a spraying control strategy is continuously optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent mines, in particular to a coal mine precise spraying control method based on point cloud segmentation and dust prediction. BACKGROUND

[0002] In the process of coal mining, a large amount of coal dust and silica dust will be generated in the operation links such as tunneling, transportation and loading. If the dust concentration exceeds the safety threshold, it will not only seriously endanger the health of the operating personnel and induce occupational diseases such as pneumoconiosis, but also increase the risk of major safety accidents such as gas explosion and coal dust explosion, becoming a key problem restricting the safety production and intelligent development of coal mines.

[0003] The existing mine generally equipped with spraying dust reduction equipment, mostly using the traditional control mode based on fixed time interval or fixed area, lacking the dynamic response ability to the actual working condition. The existing spraying system is mostly controlled by independent units, lacking the cooperative linkage mechanism, the parameters such as spraying pressure and angle are fixed, which cannot be dynamically adjusted according to the dust diffusion trend, resulting in low dust reduction efficiency, serious water resource waste, even local over-spraying or spraying blind area, which is difficult to meet the demand of modern coal mine green, safe and efficient management. SUMMARY

[0004] The purpose of the present application is to provide a coal mine precise spraying control method based on point cloud segmentation and dust prediction, aiming to improve the intelligent and fine level of coal dust control.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: the present application provides a coal mine precise spraying control method based on point cloud segmentation and dust prediction, the method comprising S1: collecting the three-dimensional point cloud data of the coal mine operation environment in real time through laser radar, using a deep learning model to perform semantic segmentation on the three-dimensional point cloud data, identifying the dust distribution area, mine air flow parameter and operation intensity information, and constructing a dynamic spatial scene semantic graph. S2: based on the dust distribution area, mine air flow parameter and operation intensity information in the dynamic spatial scene semantic graph, using an AI prediction model fusing time and space features, the dust diffusion path, concentration change trend and high risk area are predicted in real time. Among them, the AI prediction model fusing time and space features is an artificial intelligence prediction model including time sequence features and spatial distribution features, which is used to simulate the dynamic diffusion process of dust under the action of air flow. S3: according to the dust prediction result, combining the field working condition and the layout of the spraying system, the start-stop instruction, spraying direction parameter and intensity parameter of the intelligent spraying device are generated, and the dust reduction effect data is obtained, and the dust reduction effect data is fed back to the control end through the closed loop feedback mechanism, and the spraying control strategy is continuously optimized.

[0006] In some embodiments, the semantic segmentation of the three-dimensional point cloud data by using the deep learning model comprises: S11, feature extraction is performed on an original point set to obtain three-dimensional spatial coordinates, reflection intensity and surface normal vector of each point, an adaptive voxel division strategy is used to adjust a voxel scale, and a geometric feature vector is generated. The original point set is an unprocessed point cloud data set output by a laser radar. S12, a feature propagation network based on dynamic graph convolution is constructed, an adjacency graph is constructed by using a local point cloud topological relationship, and the geometric feature vector is updated by using a geometric attention weight. S13, a composite loss function is used to optimize the segmentation result. The composite loss function comprises a classification loss term and a boundary structure preservation regular term. The classification loss term indicates a difference between a predicted label and a real label, and the boundary structure preservation regular term is used to constrain spatial consistency of prediction results of adjacent points.

[0007] In some embodiments, the adaptive voxel division strategy is used to adjust the voxel scale, which means that the voxel scale is dynamically adjusted according to a point density of points in the original point set within a set radius range.

[0008] In some embodiments, the geometric attention weight is calculated based on an Euclidean distance between point pairs in the original point set, and the Euclidean distance between the point pairs is inversely proportional to a weight coefficient.

[0009] In some embodiments, the classification loss term is a cross-entropy loss. The boundary structure preservation regular term is used to preserve continuity of a semantic boundary by constraining similarity of prediction results of adjacent points.

[0010] In some embodiments, step S2 comprises: S21, a semantic scene graph embedding, a wind speed and direction vector, work intensity information and dust concentration compression features are fused to generate a spatiotemporal state vector. The spatiotemporal state vector is a comprehensive vector of spatiotemporal features of a coal mine environment at a certain moment. The semantic scene graph embedding is an embedding representation of converting a dynamic spatial scene semantic graph into a low-dimensional vector. S22, a time-gated spatial migration network is constructed, a spatial migration matrix is generated by a time-gated unit, a dust diffusion process is simulated, and a dust concentration distribution at a next moment is predicted. The time-gated spatial migration network comprises a neural network with a time-gated mechanism and is used to simulate a dust spatial migration process. The spatial migration matrix is a matrix representing migration probability of dust between different regions. S23, a trend consistency regularization loss function is used to optimize the model for prediction optimization. The loss function comprises a mean square error loss term and a diffusion trend constraint term, and the diffusion trend constraint term is used to ensure gradient consistency of a predicted dust diffusion direction and a real diffusion direction.

[0011] In some embodiments, the time-gated unit normalizes element values of the spatial migration matrix to a (0, 1) interval by using an activation function to represent migration probability of dust between regions.

[0012] In some embodiments, step S3 further comprises synergistically controlling the intelligent spraying device according to the spraying control strategy. The intelligent spraying device comprises a plurality of spraying units. Synergistically controlling the intelligent spraying device according to the spraying control strategy comprises: S31: establishing a spraying unit synergistic regulation and control model, taking spraying intensity as a control variable, constructing an optimization function containing a dust concentration control target and a resource consumption target, and solving an optimal spraying intensity parameter under a resource constraint condition. The spraying unit synergistic regulation and control model is an optimization model for coordinating working parameters of the plurality of spraying units. The resource constraint condition is a limitation condition for the intensity range of a single spraying unit and the total spraying intensity of the system. S32: dynamically adjusting a spraying pressure parameter based on a dust concentration error, and adjusting a spraying angle parameter according to a dust concentration gradient direction. The dust concentration error refers to a difference between a predicted dust concentration and a target concentration threshold. S33: introducing a deep reinforcement learning strategy, taking the current dust concentration, the wind flow parameter, and the historical spraying parameter as state input, taking the spraying intensity parameter as action output, and optimizing a decision model through a reward function. The reward function is a feedback function set according to the dust concentration reduction degree and the resource consumption condition.

[0013] In some embodiments, the optimization function aims to minimize the weighted sum of the predicted dust concentration and the spraying resource consumption.

[0014] In some embodiments, the deep reinforcement learning strategy updates the decision network parameter by minimizing the Bellman error.

[0015] Compared with the prior art, the application has the following beneficial effects: 1. The coal mine precise spraying control method based on point cloud segmentation and dust prediction provided by the embodiments of the application can accurately identify dust distribution areas, mine air flow parameters, and work intensity information by collecting three-dimensional point cloud data through a laser radar and performing semantic segmentation in combination with a deep learning model, and construct a dynamic spatial scene semantic graph. Based on the multi-dimensional information in the dynamic spatial scene semantic graph, an AI prediction model fusing spatio-temporal features is used to simulate a dust diffusion process, which can predict a dust diffusion path, a concentration change trend, and a high-risk area in real time. The model breaks through the limitations of traditional methods based on experience rules or simplified physical models, can accurately depict the nonlinear and dynamic diffusion characteristics of dust under the action of air flow, identifies potential risks in advance, provides forward-looking guidance for spraying control, and improves the initiative and scientific nature of coal mine dust control.

[0016] 2. According to the dust prediction result, the start-stop instruction, direction and intensity parameters of the intelligent spraying device are generated, and the control strategy is continuously optimized through a closed-loop feedback mechanism, so that on-demand, fixed-point and dynamic precise spraying can be realized. Compared with the existing fixed parameter and independent control spraying method, this method can realize multi-device coordinated linkage combined with the field working condition and the layout of the spraying system, reduce water resource waste while ensuring dust reduction effect, avoid local over-spraying or spraying blind area, significantly improve the efficiency and economy of spraying dust reduction, and promote the intelligent and fine upgrading of coal mine dust control. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 FIG. 1 is a schematic diagram of a coal mine precise spraying control method based on point cloud segmentation and dust prediction provided by the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] In the description of the application, it should be understood that the terms "upper", "lower", "left", "right", "front", "back", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or relative position relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements indicated must have a particular orientation, be constructed and operated in a particular orientation, therefore cannot be understood as a limitation on the present application. Unless otherwise specified, the above directional description can be flexibly set in the process of actual application, under the condition of meeting the relative position relationship shown in the drawings.

[0020] The terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, unless otherwise specified, the meaning of "multiple" is two or more.

[0021] In the description of the present application, it should be noted that unless specifically defined and limited, the terms "mounting", "connection", "connecting", "communication" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integrally connected. It can be directly connected, or indirectly connected through an intermediate medium, or the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0022] In the embodiments of the present application, the terms "comprising", "including" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, article or apparatus including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, article or apparatus. Without more limitation, the element defined by the sentence "including a…" does not exclude the presence of another identical element in the process, article or apparatus including the element.

[0023] In the embodiments of the present application, the words "exemplary" or "for example" are used to mean serving as an example or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or advantageous than other embodiments or design solutions. Rather, the use of "exemplary" or "for example" is intended to present concepts in a concrete manner.

[0024] As described in the background, there is an urgent need for an integrated technology that can realize accurate perception of working environment, dynamic prediction of dust diffusion and intelligent collaborative control of spraying system, to solve the problems of insufficient perception accuracy, prediction lag and rigid control strategy in the prior art, and to improve the intelligent and fine level of coal mine dust control.

[0025] Therefore, the embodiments of the present application provide a coal mine accurate spraying control method based on point cloud segmentation and dust prediction, for example, as shown in Figure 1 The method comprises: Perception and modeling of working environment S1: collecting three-dimensional point cloud data of coal mine working environment in real time through laser radar, performing semantic segmentation on the three-dimensional point cloud data by using a deep learning model, identifying dust distribution area, mine air flow parameter and work intensity information, and constructing a dynamic space scene semantic graph.

[0026] Among them, the three-dimensional point cloud data refers to the data set containing the three-dimensional space coordinates and reflection characteristics of each point in the coal mine working environment obtained by the laser radar emitting laser beams and receiving reflection signals. The dynamic space scene semantic graph refers to a space scene model that integrates semantic information such as dust distribution, air flow parameter and work intensity, and is dynamically updated with the working environment.

[0027] In some embodiments, a deep learning model is used to perform semantic segmentation on 3D point cloud data, including S11: Extract features from the original point set to obtain the three-dimensional spatial coordinates, reflection intensity and surface normal vector of each point, use an adaptive voxel partitioning strategy to adjust the voxel scale, and generate a geometric feature vector.

[0028] The original point set is the unprocessed point cloud data set output by the lidar. A geometric feature vector refers to a feature vector containing geometric attributes such as the point's three-dimensional coordinates, reflection intensity, and surface normal. In some embodiments, using an adaptive voxel partitioning strategy to adjust voxel scale involves dynamically adjusting the voxel scale based on the point density within a set radius within the original point set to balance the impact of uneven point cloud density on feature extraction.

[0029] As a possible implementation method, in order to improve the expression ability of point cloud data in complex coal mine scenes, the original point set is first Perform preprocessing; each point Expressed as three-dimensional space coordinates , and its reflection intensity and the surface normal vector To avoid interference from uneven density on learning, an adaptive voxel partitioning strategy is adopted to dynamically adjust the scale of the voxel where each point is located. , defined as follows: in, for point In radius The point density inside Indicated by point The spatial volume of a three-dimensional spherical neighborhood with a center and a radius of r, To tune the voxel scale hyperparameter, is a small constant that prevents division by zero, for point The points around ; the geometric feature vector of the extracted point is: Used for deep semantic modeling input to ensure that both structural and material information are encoded.

[0030] S12: Construct a feature propagation network based on dynamic graph convolution, use the local point cloud topological relationship to build an adjacency graph, and update the geometric feature vector through geometric attention weights.

[0031] wherein the dynamic graph convolution refers to a graph convolution operation of dynamically adjusting a convolution kernel parameter according to a local point cloud topological structure; and the geometric attention weight refers to a weight coefficient for representing importance of a neighborhood point, which is calculated based on a spatial distance between point pairs. In some embodiments, the geometric attention weight is calculated based on an Euclidean distance between point pairs in the original point set, and the Euclidean distance of the point pair is inversely proportional to the weight coefficient.

[0032] As a possible implementation, in the deep semantic modeling stage, a feature propagation structure based on dynamic graph convolution is constructed, and an adjacency graph is dynamically constructed using local point cloud topological information, and the update formula is: wherein, is an embedding representation of a point in the i-th layer, is a feature of an adjacent point in the i-th layer, is a feature of an adjacent point in the i-th layer, is a shared perception machine, is a local neighborhood point set of a point, is a geometric attention weight between point pairs, used to enhance geometric correlation modeling, and is defined as follows: is an Euclidean distance between point pairs, wherein, is an Euclidean distance between point pairs, is a distance decay control parameter to prevent interference of long-distance point pairs.

[0033] S13: The segmentation result is optimized by using a composite loss function.

[0034] wherein the composite loss function includes a classification loss term and a boundary structure preservation regular term. The classification loss term indicates a difference between a predicted label and a true label, and the boundary structure preservation regular term is used to constrain spatial consistency of prediction results of adjacent points.

[0035] In some embodiments, the classification loss term is a cross-entropy loss. The boundary structure preservation regular term maintains the continuity of semantic boundaries by constraining the similarity of prediction results of adjacent points.

[0036] More specifically, to improve the semantic accuracy and spatial consistency of the segmentation result, a composite loss function is designed, which is composed of a classification loss and a boundary structure preservation regular term: wherein is a standard cross-entropy loss, which measures a difference between a predicted label and a true label, and the boundary structure preservation regular term is used to constrain spatial consistency of prediction results of adjacent points.​​​​​​ the difference between the real label ; the second term is a spatially smooth regularization term, which is used to constrain the prediction results of adjacent points and are similar, and the semantic boundary is continuous, and is a weight factor of the regularization term; is a weight factor of the regularization term.

[0037] S2: Based on the dust distribution area in the dynamic spatial scene semantic graph, the mine air flow parameters and the work intensity information, an AI prediction model fusing spatio-temporal features is used to predict the dust diffusion path, the concentration change trend and the high-risk area in real time.

[0038] The purpose of step S2 is to dynamically predict the dust diffusion, including dynamically simulating and predicting the dust diffusion path, the concentration change and the high-risk area, and providing forward-looking pollution trend evaluation. The AI prediction model fusing spatio-temporal features is an artificial intelligence prediction model including time sequence features and spatial distribution features, which is used to simulate the dynamic diffusion process of dust under the action of air flow.

[0039] In some embodiments, step S2 includes: S21: Fusing semantic scene graph embedding, wind speed and direction vector, work intensity information and dust concentration compression features to generate a spatio-temporal state vector. The spatio-temporal state vector is a comprehensive vector of the spatio-temporal features of the coal mine environment at a certain moment. The semantic scene graph embedding is an embedding representation of converting the dynamic spatial scene semantic graph into a low-dimensional vector.

[0040] An exemplary spatio-temporal state vector is constructed by fusing multiple source input features , which describes the coal mine environment state at moment : wherein, is the semantic scene graph embedding, representing the point cloud semantic distribution and boundary structure; is the wind speed and direction vector, used for modeling the diffusion driving force; is the equipment and personnel work intensity information; is the low-dimensional compression representation of the current dust concentration distribution graph; is the vector splicing operation, used for unified modeling input; the state vector comprehensively considers the spatial layout, physical driving and pollution source intensity, and is the basis for the prediction of the AI prediction model.

[0041] ​S22: Construct a time-gated spatial migration network. This network generates a spatial migration matrix using time-gated units to simulate the dust diffusion process and predict the dust concentration distribution at the next moment. The time-gated spatial migration network includes a neural network with a time-gating mechanism for simulating the spatial migration of dust. The spatial migration matrix represents the probability of dust migrating between different regions.

[0042] In some embodiments, the time gating unit normalizes the element values ​​of the spatial migration matrix to the (0, 1) interval through an activation function to represent the probability of dust migration between regions.

[0043] Time-Gated Spatial Migration Network (TGSM-Net); its core is to introduce a time-gated unit to control the spatial migration matrix , dynamic modeling diffusion weight: in, It's prediction time The diffusion migration matrix represents the pollution from The probability of movement between regions; Is the Sigmoid function, used to normalize the migration weight to interval; It is a learnable parameter of the AI ​​prediction model, which is used to transform the state vector Mapping is a diffusion structure; Predicting the next moment The dust concentration distribution vector is obtained by the time gating mechanism, and the spatial migration matrix is ​​dynamically adjusted to simulate the dust diffusion process under complex wind flow and environment.

[0044] S23: A trend consistency regularized loss function optimization model is used for prediction optimization. The loss function includes a mean square error loss term and a diffusion trend constraint term. The diffusion trend constraint term is used to ensure the gradient consistency between the predicted dust diffusion direction and the actual diffusion direction.

[0045] The setting of loss function can effectively enhance the sensitivity of AI prediction model to diffusion trend. For example, the trend consistency regularization loss is proposed. , encouraging the prediction results of the AI ​​prediction model to be consistent with the actual diffusion direction in gradient: in, is the standard mean square error loss, and the supervised predicted concentration is consistent with the true concentration, For the moment The actual dust concentration; and respectively represent the predicted and real concentration gradient at spatial position reflecting the diffusion direction; is a trade-off term, used to adjust the influence of the trend regularizer, is the number of intelligent spraying devices; this loss strengthens the alignment of the diffusion structure on the basis of regression accuracy, making the AI prediction model robust and practical in predicting the evolution path of pollution hotspots.

[0046] S3: According to the dust prediction result, combined with the field working conditions and the layout of the spraying system, the start-stop instruction, spraying direction parameter and intensity parameter of the intelligent spraying device are generated, and the dust reduction effect data is obtained. Through the closed-loop feedback mechanism, the dust reduction effect data is fed back to the control end, and the spraying control strategy is continuously optimized.

[0047] In some embodiments, step S3 further comprises cooperatively controlling the intelligent spraying device according to the spraying control strategy. Cooperatively controlling the intelligent spraying device according to the spraying control strategy comprises: S31: Establish a spraying unit cooperative control model, taking the spraying intensity as the control variable, construct an optimization function containing the dust concentration control target and the resource consumption target, and solve the optimal spraying intensity parameter under the condition of meeting the resource constraint condition.

[0048] Wherein, the spraying unit cooperative control model is an optimization model for coordinating the working parameters of multiple spraying units. The resource constraint condition is the limiting condition for the intensity range of a single spraying unit and the total spraying intensity of the system. In some embodiments, the objective of the optimization function is to minimize the weighted sum of the predicted dust concentration and the spraying resource consumption.

[0049] The intelligent spraying device includes multiple spraying units. For example, a spraying unit cooperative control model is established, and for an intelligent spraying device, the spraying intensity of each spraying unit at time is defined as the control variable ∈[0,1], which is used to adjust the opening degree of the spraying valve, 0 represents closed, and 1 represents the maximum spraying intensity; the objective is to reduce resource consumption while meeting the dust control demand, and the following optimization objective function is constructed : Wherein, is the predicted dust concentration distribution vector of the th region at the next time in step S2; is the spraying intensity control variable of the current th spraying unit; , are the weighted coefficients of pollution control and resource consumption; is the comprehensive objective function value, the smaller the better; meanwhile, the resource constraint condition is introduced: wherein, is the maximum total spray intensity that the system can withstand, limiting excessive spraying and resource waste; the spray unit cooperative control model can be solved online in real time, outputting the optimal spray distribution scheme.

[0050] S32: dynamically adjusting the spray pressure parameter based on the dust concentration error, and adjusting the spray angle parameter according to the dust concentration gradient direction. The dust concentration error refers to the difference between the predicted dust concentration and the target concentration threshold.

[0051] Considering the dynamic changes of mine operation, the dust generation and diffusion have nonlinear time-varying characteristics, therefore, an adaptive control mechanism is introduced for the pressure and direction parameters in the spray system to achieve fine control. For example, the spray pressure of the first spray unit is defined as , which affects the spray distance and coverage range; the spray angle is which determines the spray direction and area; the target concentration threshold is set to , then the concentration error of the first area is: An update rule with proportional-differential control is adopted: wherein, is the proportional and differential gain coefficient of the spray pressure, adjusting the response rate and stability; is the gradient direction of the dust concentration in space, indicating the diffusion trend; sign(·) is a sign function, used to adjust the spray direction to tend to the high concentration area; is the angle adjustment step; this mechanism can real-time fine-tune the spray physical parameters according to the current prediction, achieving double dynamic response.

[0052] S33: introducing a deep reinforcement learning strategy, taking the current dust concentration, wind flow parameters and historical spray parameters as state inputs, and taking the spray intensity parameter as action output, and optimizing the decision model through a reward function.

[0053] wherein, the reward function refers to a feedback function set according to the dust concentration reduction degree and resource consumption. In some embodiments, the deep reinforcement learning strategy updates the decision network parameters by minimizing the Bellman error.

[0054] For example, under dynamic working conditions and uncertain environment, a cooperative control strategy based on deep reinforcement learning is introduced to improve the intelligent decision-making ability of the system, and the system state is defined as:​​ wherein, is the current predicted dust concentration of the region; is the current wind speed and direction vector; is the previous spray control decision; defines the action is: i.e. the control instruction of each spray unit, the immediate reward function is designed as: wherein, is the dust residue penalty coefficient; is the spray resource penalty coefficient; the reward value The greater the value, the better the system, the goal is to minimize the concentration and resource consumption; using a deep reinforcement learning framework, the Bellman error is used to update the policy network loss function : wherein, is a discount factor, used to weigh the importance of the current reward and the long-term reward in the future, and the value is between 0-1, is the current network parameter; is the expectation of the state , action , reward , next state sampled in experience replay, is the target reinforcement learning network parameter; is the action value under the current policy, is the maximum action value under all possible actions in the next state , representing the future expected return under the optimal policy; through reinforcement learning strategy, the spray behavior can be learned through continuous interaction, and the multi-unit linkage control can be realized, which enhances the adaptability and robustness of the system in complex scenarios.

[0055] The present application aims at the key problems of high dust concentration, difficult to control diffusion, and lagging spray response in coal mine operating environment, and proposes an intelligent dust reduction method combining real-time segmentation of laser radar point cloud, AI prediction of dust diffusion, and cooperative control of intelligent spray device, and constructs an integrated system architecture of "space perception-dynamic prediction-intelligent decision-closed loop control".

[0056] Through laser radar to collect three-dimensional point cloud data, combined with deep learning model for semantic segmentation, accurate identification of operation area, equipment distribution and dust cloud position, real-time perception of mine space structure and pollution situation (corresponding to step S1 of operation environment perception and modeling); Further introduce the spatio-temporal prediction model of fusion wind speed, wind direction, operation intensity and other multi-source data, based on deep neural network to dynamically simulate and risk warning of dust concentration evolution trend (corresponding to step S2 of dust diffusion dynamic prediction); Finally, design the spray collaborative control strategy based on optimization algorithm and reinforcement learning, intelligent adjustment of the opening intensity, angle and timing of the spray unit, realize the linkage response among multiple devices and dynamic optimal allocation of spray resources (corresponding to step S3 of intelligent spray collaborative control).

[0057] The method provided by the embodiments of the present application has high real-time, adaptability and intelligence, which significantly improves the spatial coverage efficiency and resource utilization of dust reduction. In actual coal mine application, it can effectively reduce the dust concentration peak value and over-standard time, reduce blind area spraying and resource waste, improve mine operation safety and environmental quality, and provide key technical support for promoting the intelligent upgrading and green low-carbon transformation of the coal mine ventilation and dust removal system.

[0058] In the description of the present specification, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0059] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A coal mine precision spray control method based on point cloud segmentation and dust prediction, characterized in that: include: S1: Use LiDAR to collect 3D point cloud data of the coal mine operating environment in real time, use a deep learning model to perform semantic segmentation on the 3D point cloud data, identify dust distribution areas, mine airflow parameters, and operating intensity information, and construct a dynamic spatial scene semantic map; S2: Based on the dust distribution area, mine airflow parameters, and operation intensity information in the dynamic spatial scene semantic map, an AI prediction model integrating spatiotemporal features is used to perform real-time predictions of dust diffusion paths, concentration change trends, and high-risk areas. The AI ​​prediction model integrating spatiotemporal features is an artificial intelligence prediction model that includes time series features and spatial distribution features and is used to simulate the dynamic diffusion process of dust under the action of wind flow. S3: Based on the dust prediction results, combined with the on-site working conditions and the layout of the spray system, the start and stop instructions, spray direction parameters and intensity parameters of the intelligent spray device are generated, and the dust reduction effect data is obtained. The dust reduction effect data is transmitted back to the control end through a closed-loop feedback mechanism to continuously optimize the spray control strategy.

2. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 1 is characterized in that: The performing semantic segmentation on the three-dimensional point cloud data using a deep learning model includes: S11: extracting features from the original point set to obtain the three-dimensional spatial coordinates, reflection intensity, and surface normal vector of each point, adjusting the voxel scale using an adaptive voxel partitioning strategy, and generating a geometric feature vector; wherein the original point set is a set of unprocessed point cloud data output by the lidar; S12: Construct a feature propagation network based on dynamic graph convolution, build an adjacency graph using the local point cloud topological relationship, and update the geometric feature vector through geometric attention weights; S13: Optimize the segmentation results using a composite loss function; wherein the composite loss function includes a classification loss term and a boundary structure preservation regularization term; the classification loss term indicates the difference between the predicted label and the true label, and the boundary structure preservation regularization term is used to constrain the spatial consistency of the prediction results of adjacent points.

3. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 2 is characterized in that: The adopting of the adaptive voxel division strategy to adjust the voxel scale refers to dynamically adjusting the voxel scale according to the point density of the points in the original point set within a set radius.

4. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 2 is characterized in that: The geometric attention weight is calculated based on the Euclidean distance between point pairs in the original point set, and the Euclidean distance between the point pairs is inversely proportional to the weight coefficient.

5. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 2 is characterized in that: The classification loss term is a cross entropy loss; the boundary structure preserving regularization term maintains the continuity of the semantic boundary by constraining the similarity of the prediction results of adjacent points.

6. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 1 is characterized in that: Step S2 includes: S21: Fusing the semantic scene graph embedding, wind speed and direction vectors, operation intensity information, and dust concentration compression features to generate a spatiotemporal state vector; wherein the spatiotemporal state vector is a comprehensive vector of spatiotemporal characteristics of the coal mine environment at a certain moment; the semantic scene graph embedding is an embedding representation that converts the dynamic spatial scene semantic graph into a low-dimensional vector; S22: Constructing a time-gated spatial migration network, generating a spatial migration matrix through a time-gated unit, simulating the dust diffusion process, and predicting the dust concentration distribution at the next moment; wherein the time-gated spatial migration network includes a neural network with a time-gating mechanism for simulating the dust spatial migration process; the spatial migration matrix is ​​a matrix representing the probability of dust migration between different areas; S23: Use the trend consistency regularized loss function optimization model to perform prediction optimization; the loss function includes a mean square error loss term and a diffusion trend constraint term, and the diffusion trend constraint term is used to ensure the gradient consistency between the predicted dust diffusion direction and the actual diffusion direction.

7. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 6 is characterized in that: The time gating unit normalizes the element values ​​of the spatial migration matrix to the (0, 1) interval through an activation function to represent the probability of dust migration between regions.

8. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 1 is characterized in that: The step S3 further includes collaboratively controlling the intelligent spray device according to the spray control strategy; the intelligent spray device includes a plurality of spray units; The collaborative control of the intelligent spray device according to the spray control strategy includes: S31: Establish a spray unit coordinated control model, using spray intensity as a control variable, construct an optimization function that includes a dust concentration control target and a resource consumption target, and solve for the optimal spray intensity parameter while satisfying resource constraints; wherein the spray unit coordinated control model is an optimization model for coordinating the operating parameters of multiple spray units; and the resource constraints are restrictions on the intensity range of a single spray unit and the total spray intensity of the system; S32: Dynamically adjust the spray pressure parameter based on the dust concentration error and adjust the spray angle parameter according to the direction of the dust concentration gradient; where the dust concentration error refers to the difference between the predicted dust concentration and the target concentration threshold; S33: Introduce a deep reinforcement learning strategy, take the current dust concentration, wind flow parameters and historical spray parameters as state input, take the spray intensity parameters as action output, and optimize the decision model through a reward function; wherein, the reward function refers to a feedback function set according to the degree of dust concentration reduction and resource consumption.

9. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 8, characterized in that: The goal of the optimization function is to minimize the weighted sum of the predicted dust concentration and the spray resource consumption.

10. The method for coal mine precision spray control based on point cloud segmentation and dust prediction according to claim 8, characterized in that: The deep reinforcement learning strategy updates the decision network parameters by minimizing the Bellman error.

Images