Intelligent laser coal checking system and method

Abstract

The invention relates to the technical field of computer programming languages, and particularly discloses an intelligent laser coal checking system and method. The system comprises a closed-loop architecture formed by an environment adaptive data acquisition module, a layered data transmission module, a real-time modeling analysis module and an anomaly detection and early warning module, and a genetic algorithm driving engine collaborative optimization. Dust concentration and holder parameters are collected in real time through a distributed sensor network, and redundant lens switching and dynamic compensation are triggered; the anomaly detection module constructs Bayesian network associated multi-source data features, and performs root cause analysis by adopting Monte Carlo tree search; and the closed-loop optimization module realizes cross-coal-yard differential privacy parameter migration through a federal evolution mechanism, and dynamically adjusts the weight of a target function to form a self-learning closed loop. According to the method, the problems of equipment failure, modeling delay and early warning false alarm and missing alarm in a high-dust environment are effectively solved, and the real-time performance of coal pile monitoring and the robustness of the system are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data processing, and in particular relates to an intelligent laser coal processing system and method. Background Art

[0002] The intelligent laser coal-handling system is a digital coal yard management solution based on laser scanning technology, primarily used for real-time monitoring of the volume, weight, and distribution of coal piles. The intelligent laser coal-handling system uses a fixed laser scanner and a rotating pan-tilt head installed on top of the coal shed, combined with a three-dimensional modeling algorithm, to fully automatically scan the coal pile, generate high-precision three-dimensional point cloud data, and implement zoning and layered management of the coal pile through a visual interface. The core technologies of the intelligent laser coal-handling system include laser ranging, synchronous data transmission, three-dimensional modeling, and data analysis. It can replace traditional manual measurement or mechanical coal-handling methods, significantly improving coal-handling efficiency and accuracy. In addition, the system supports remote control using a B / S architecture, allowing managers to view coal yard dynamics in real time through a browser, providing data support for coal inventory management, coal blending optimization, and safety monitoring.

[0003] Coal yards are subject to high dust concentrations and significant temperature and humidity fluctuations, requiring equipment to meet explosion-proof and corrosion-resistant requirements. However, existing systems are prone to problems such as lens contamination and motor jamming under extreme operating conditions, impacting long-term stable operation. Coal piles experience frequent dynamic changes, and the existing model reconstruction algorithm still exhibits delays in the "1-minute rough sweep, 6-minute fine sweep" mode. Under these extreme conditions, the existing system is prone to lens contamination and motor jamming, making it difficult to meet high real-time requirements and hindering the timeliness of abnormality warnings provided by intelligent coal panning technology. Summary of the Invention

[0004] In response to the technical pain points existing in the above-mentioned existing technologies, the present invention provides an intelligent laser coal panning system, method and device to solve the problem that when the existing system is prone to lens contamination, motor jamming and other problems under extreme working conditions, it is difficult to meet high real-time requirements, which affects the timeliness of abnormal warning of intelligent coal panning technology.

[0005] In a first aspect, the present invention provides an intelligent laser coal mining system, comprising an environment-adaptive data acquisition module, a hierarchical data transmission module, a real-time modeling and analysis module, an anomaly detection and early warning module, all sequentially connected via data and control flows, as well as a closed-loop optimization module and a genetic algorithm-driven engine that interacts with each module in a two-way manner. The environmental adaptive data acquisition module collects dust concentration and pan / tilt motion parameters in real time through a distributed sensor network, triggers redundant lens switching and dynamic compensation according to preset thresholds, and outputs the original data stream to the layered data transmission module; After receiving the original data stream, the layered data transmission module generates standardized feature data through edge preprocessing, optimizes feature selection and compression strategies based on genetic algorithms, and transmits the compressed valid data to the real-time modeling and analysis module; The real-time modeling and analysis module constructs a reference three-dimensional grid based on an incremental modeling engine, combines the characteristic data provided by the hierarchical data transmission module, and uses a genetic algorithm to optimize local reconstruction parameters to generate a real-time coal pile model; The anomaly detection and warning module receives the real-time coal pile model and sensor data, associates multi-source data features through the knowledge graph, and triggers a hierarchical warning signal to be fed back to the closed-loop optimization module; The closed-loop optimization module works in conjunction with the genetic algorithm driving engine to dynamically optimize the parameters of each module based on early warning signals and system performance indicators.

[0006] Furthermore, in the intelligent laser coal processing system of the present invention, the genetic algorithm driving engine includes: The parameter optimization unit, based on the chromosome encoding mechanism of the genetic algorithm, jointly optimizes the scanning frequency of the data acquisition module, the accuracy level of the modeling module, and the compression ratio of the transmission module to generate a collaborative parameter set; The feature selection unit screens the optimal combination of geometric features, temporal features, and environmental features based on the gene pool to generate a feature compression weight matrix for the hierarchical transmission module; The path planning unit generates an adaptive scanning path based on the deformation detection results of the real-time modeling module and dynamically adjusts the pan-tilt motion trajectory of the data acquisition module; The model evolution unit uses the neural network architecture to search and optimize the anomaly detection model, and feeds the evolved model parameters back to the early warning module.

[0007] Furthermore, in the intelligent laser coal processing system of the present invention, the layered data transmission module includes a cascaded data cleaning unit, a dynamic compression unit, and an adaptive sharding unit: The data cleaning unit receives the raw data stream from the data acquisition module and removes outliers based on the preset dust concentration threshold and statistical filtering algorithm; The dynamic compression unit adopts a layered compression strategy for the cleaned data and optimizes the space division and encoding parameters based on the genetic algorithm; The adaptive sharding unit dynamically adjusts the data sharding granularity and adds forward error correction code based on the network delay prediction value fed back by the real-time modeling module.

[0008] Furthermore, in the intelligent laser coal mining system of the present invention, the real-time modeling and analysis module includes iterative optimization: The difference detection unit locates the changed area reported by the data acquisition module based on the spatial hash table optimized by the genetic algorithm; The multi-granularity modeling unit generates a multi-resolution coal pile model by synchronously optimizing octree voxelization and Poisson surface reconstruction parameters through dynamic chromosome encoding; The trend prediction unit uses a non-dominated sorting genetic algorithm to balance modeling error and computing resource consumption, and outputs the prediction results to the anomaly detection module.

[0009] Furthermore, in the intelligent laser coal processing system of the present invention, the closed-loop optimization module is implemented through a feedback loop: The federated evolution unit collects anomaly detection data from each coal yard node and performs chromosome migration through a differential privacy protocol; Dynamic weight allocation unit, which dynamically adjusts the objective function weight of the detection model according to the false positive or false negative statistics of the early warning module; The feedback learning unit injects the optimized parameter set back into the genetic algorithm driving engine to form a closed loop of parameter evolution.

[0010] In a second aspect, the present invention provides an intelligent laser coal processing method, which is applied to the intelligent laser coal processing system, comprising: S1: Real-time environmental data is acquired through the distributed sensors of the data acquisition module to trigger redundant equipment switching and dynamic compensation; S2: Input the data collected by S1 into the hierarchical transmission module and use genetic algorithm to optimize feature selection and data compression strategy; S3: Based on the feature data processed by S2, a real-time 3D model is generated through incremental modeling and multi-granularity reconstruction in the modeling and analysis module; S4: Input the modeling results of S3 into the anomaly detection module, combine the knowledge graph and evolutionary model to identify abnormal features and trigger an alert; S5: Based on the early warning feedback from S4, the closed-loop optimization module dynamically adjusts the system parameters and updates the genetic algorithm evolution strategy.

[0011] Furthermore, in the intelligent laser coal processing method of the present invention, S2 specifically includes: S21: Receives the dust concentration and gimbal motion parameter data transmitted by S1, and screens the optimal combination of geometric features and environmental features through the gene pool of the genetic algorithm; S22: Use a hierarchical compression strategy to perform KD-tree partitioning, PCA dimensionality reduction, and hybrid encoding on the selected features; S23: Dynamically adjust the compression ratio and sharding granularity based on the computing resource status fed back by the S3 module.

[0012] Furthermore, in the intelligent laser coal processing method of the present invention, S3 includes: S31: Build a spatial hash table based on the pre-processed data in S2 to locate the deformation area reported by the data acquisition module; S32: Simultaneous optimization of voxelization resolution and surface reconstruction parameters via dynamic chromosome encoding; S33: Use the NSGA-II algorithm to balance modeling accuracy and resource consumption, and output the optimized model to S4.

[0013] Furthermore, in the intelligent laser coal processing method of the present invention, the step S4 executes: S41: Establish a Bayesian network association model of equipment status, environmental parameters and modeling errors; S42: Monte Carlo tree search for abnormal root cause analysis based on the deformation prediction results of S3; S43: Update the detection model parameters using the false positive data fed back from S5.

[0014] Furthermore, in the intelligent laser coal processing method of the present invention, S5 includes: S51: Collect performance data of each module every 24 hours and execute genetic algorithm evolution iteration; S52: Using differential privacy protocol to achieve cross-coal yard model parameter migration; S53: Use the false positive and negative data generated by S4 as the basis for fitness evaluation and reversely optimize the feature selection strategy.

[0015] Beneficial effects of the present invention: The present invention solves the problem of sensor failure in high-dust environments through the redundant lens switching and dynamic compensation technology of the environmental adaptive data acquisition module, and combines the genetic algorithm optimization compression strategy and adaptive fragmentation mechanism of the layered data transmission module to significantly improve the data transmission efficiency and stability; the real-time modeling and analysis module adopts incremental modeling and multi-granularity reconstruction technology to greatly reduce computing resource consumption while ensuring the accuracy of the three-dimensional model; the anomaly detection and early warning module associates multi-source data features through the knowledge graph to achieve accurate hierarchical early warning; the closed-loop optimization module is based on the dynamic parameter adjustment and federated evolution mechanism of the genetic algorithm-driven engine to continuously optimize the overall system performance, ultimately achieving high real-time, high robustness and cross-site collaborative optimization capabilities of coal pile monitoring, effectively improving the intelligence level of coal inventory management and safety early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a timing diagram of the intelligent laser coal processing system provided by the technical solution of the present invention. DETAILED DESCRIPTION

[0017] An embodiment of the present invention will be further described below with reference to the accompanying drawings.

[0018] See also Figure 1In a first aspect, the present invention provides an intelligent laser coal mining system, comprising an environment-adaptive data acquisition module, a hierarchical data transmission module, a real-time modeling and analysis module, an anomaly detection and early warning module, which are sequentially connected via data flow and control flow, as well as a closed-loop optimization module and a genetic algorithm driving engine that bidirectionally interact with each module. The environmental adaptive data acquisition module collects dust concentration and pan / tilt motion parameters in real time through a distributed sensor network, triggers redundant lens switching and dynamic compensation according to preset thresholds, and outputs the original data stream to the layered data transmission module; After receiving the original data stream, the layered data transmission module generates standardized feature data through edge preprocessing, optimizes feature selection and compression strategies based on genetic algorithms, and transmits the compressed valid data to the real-time modeling and analysis module; The real-time modeling and analysis module constructs a reference three-dimensional grid based on an incremental modeling engine, combines the characteristic data provided by the hierarchical data transmission module, and uses a genetic algorithm to optimize local reconstruction parameters to generate a real-time coal pile model; The anomaly detection and warning module receives the real-time coal pile model and sensor data, associates multi-source data features through the knowledge graph, and triggers a hierarchical warning signal to be fed back to the closed-loop optimization module; The closed-loop optimization module works in conjunction with the genetic algorithm driving engine to dynamically optimize the parameters of each module based on early warning signals and system performance indicators.

[0019] Environmental adaptive data acquisition module: Sensor network configuration: The distributed sensor network consists of lidar, infrared ranging sensor and dust concentration sensor. The lidar is linked with the gimbal mechanical structure to achieve multi-view scanning through gimbal rotation angle control.

[0020] Redundant switching logic: When the dust concentration sensor detection value exceeds the preset threshold, the backup lens switching command is triggered, and the gimbal attitude adjustment (pitch angle compensation) is used to offset the scanning blind spot caused by lens switching.

[0021] Dynamic compensation mechanism: Based on the real-time changes in dust concentration gradient, the lidar scanning frequency and infrared sensor sampling interval are adaptively adjusted to ensure the continuity of data collection.

[0022] This module utilizes multi-sensor collaborative data collection to address the vulnerability of single sensors to failure in high-dust environments. Redundant lens switching and dynamic compensation form a closed-loop control system to prevent data interruptions caused by equipment contamination or environmental interference. For example, if the primary lens's scanning accuracy decreases due to dust coverage, the system automatically switches to the backup lens and uses pan / tilt angle compensation to ensure a complete scanning range, providing a stable data source for subsequent modeling.

[0023] Hierarchical data transmission module: Edge preprocessing process: Data cleaning: Remove noise caused by dust interference based on statistical outlier detection algorithms (such as DBSCAN clustering); Feature standardization: Coordinate system alignment and normalization are performed on the lidar point cloud data to generate a uniform-scale feature vector.

[0024] Dynamic compression strategy: A hierarchical compression mechanism is adopted. The first layer performs regional clustering based on spatial grid division (KD-Tree). The second layer reduces feature dimensions through principal component analysis (PCA). Finally, Huffman coding is combined to generate compressed data packets.

[0025] Genetic algorithm optimization: The feature selection problem is converted into chromosome encoding (such as binary genes representing feature weights), and the balance between compression rate and data fidelity is evaluated through the fitness function to iteratively generate the optimal feature subset.

[0026] This module addresses the issue of data transmission delays for massive point clouds through a three-stage process: cleaning, standardization, and layered compression. A genetic algorithm optimizes feature selection, eliminating the subjectivity of manually setting compression strategies. For example, high-precision data is prioritized at the edges of coal piles (where geometric features are prominent), while a high compression ratio is applied to flat areas (where features are limited). This achieves a dynamic balance between data transmission efficiency and modeling accuracy.

[0027] Real-time modeling and analysis module: Incremental modeling engine: Construct a benchmark 3D grid based on a historical coal pile model; Use the spatial hash table to quickly locate the difference between the newly added scan data and the baseline grid; Use a local surface reconstruction algorithm (such as Poisson reconstruction) to update the mesh vertices in the difference area.

[0028] Genetic algorithm parameter optimization: Parameters such as voxel resolution and surface smoothness are encoded into chromosomes, and the optimal parameter combination is screened through multi-generation evolution using modeling error and computation time as the dual objective function.

[0029] Multi-granularity modeling unit: Generates coal pile models of different resolutions according to anomaly detection requirements (such as a global low-precision model for rapid early warning and a local high-precision model for root cause analysis).

[0030] This module utilizes a "baseline mesh + local update" model to avoid the resource consumption of full modeling. A genetic algorithm dynamically optimizes parameters, addressing the poor adaptability of traditional modeling methods to complex coal pile configurations. For example, when a coal pile surface collapses, the system prioritizes high-precision reconstruction of the collapsed area while maintaining low-precision representations of other areas, significantly improving modeling efficiency.

[0031] Anomaly detection and early warning module: Knowledge graph construction: Device status (such as gimbal motor current), environmental parameters (such as temperature and humidity), and modeling errors (such as surface fitting residuals) are constructed as triple relationships to form a dynamically updated knowledge graph.

[0032] Graded early warning mechanism: Level 1 warning: When local deformation is detected but does not exceed the safety threshold, a visual interface annotation is triggered; Secondary warning: When the deformation area expands or the parameters of the related equipment are abnormal, the alarm information is pushed to the management terminal; Level 3 warning: When landslide risk characteristics are detected, the linkage control module will urgently stop the surrounding equipment.

[0033] Root cause analysis: Monte Carlo tree search (MCTS) is used to simulate the impact paths of multiple factors and locate the root causes of abnormalities (such as sudden changes in dust concentration leading to sensor failure, or transmission delays causing modeling deviations).

[0034] This module addresses the high false alarm rate and delayed response times of traditional early warning systems by correlating multi-source data and implementing a hierarchical response mechanism. For example, when the knowledge graph detects a correlation between an abnormal increase in gimbal motor current and a sudden increase in local modeling error, it can infer that a gimbal mechanical jam is the root cause of the anomaly and trigger targeted maintenance instructions.

[0035] Closed-loop optimization module and genetic algorithm driving engine: Dynamic parameter optimization: Every 24 hours, the operating indicators of each module (such as data compression rate, modeling error, and warning response time) are collected to generate chromosome populations, and a new generation of parameter sets is evolved through crossover and mutation operations.

[0036] Federated evolution mechanism: The closed-loop optimization modules of multiple coal yard nodes share chromosome fragments (such as feature selection weights and modeling smoothing coefficients) through differential privacy technology to achieve cross-site collaborative evolution.

[0037] Feedback learning path: Inject false positive / missing negative data back into the genetic algorithm fitness function, reduce the weight priority of similar feature combinations, and continuously optimize the system robustness.

[0038] This module addresses the poor scenario adaptability caused by fixed parameters in traditional systems through a "data-driven evolution + cross-site collaboration" mechanism. For example, if sensor sensitivity at a coal yard decreases due to seasonal humidity fluctuations, the optimized parameters can be quickly migrated to other sites through federated evolution, preventing the recurrence of the same failure.

[0039] Overall logical summary of the technical solution: Data flow closed loop: A complete closed loop is formed from data collection → transmission → modeling → detection → optimization, and each module dynamically adjusts its operating strategy through two-way feedback.

[0040] The core role of genetic algorithm: Optimize feature selection and compression strategies at the transport layer; Balance accuracy and efficiency at the modeling level; Realize parameter self-learning and cross-site migration in the optimization layer.

[0041] Abnormal handling logic: Through the four-level mechanism of "real-time detection - graded warning - root cause analysis - parameter optimization", passive response is transformed into active prevention.

[0042] Based on the above implementation methods, those skilled in the art can implement system deployment in combination with specific hardware platforms (such as industrial-grade lidar and edge computing gateways), and can understand the interaction between modules and the technical effects without creative work.

[0043] Specifically, the intelligent laser coal processing system of the present invention, the genetic algorithm driving engine includes: The parameter optimization unit, based on the chromosome encoding mechanism of the genetic algorithm, jointly optimizes the scanning frequency of the data acquisition module, the accuracy level of the modeling module, and the compression ratio of the transmission module to generate a collaborative parameter set; The feature selection unit screens the optimal combination of geometric features, temporal features, and environmental features based on the gene pool to generate a feature compression weight matrix for the hierarchical transmission module; The path planning unit generates an adaptive scanning path based on the deformation detection results of the real-time modeling module and dynamically adjusts the pan-tilt motion trajectory of the data acquisition module; The model evolution unit uses the neural network architecture to search and optimize the anomaly detection model, and feeds the evolved model parameters back to the early warning module.

[0044] Parameter optimization unit: Chromosome encoding mechanism: The data acquisition module's scanning frequency (e.g., 10Hz-50Hz), the modeling module's accuracy level (e.g., voxel resolution 0.1µm-1µm), and the transmission module's compression ratio (e.g., 20%-80%) are mapped to chromosome gene bits. Each gene bit is encoded using a real number to represent the parameter's value range. For example, gene bit 1 represents the scanning frequency, and gene bit 2 represents the modeling accuracy.

[0045] Fitness function design: Data acquisition efficiency (scanning coverage per unit time), modeling error (point cloud fitting residual), and transmission bandwidth occupancy are used as evaluation indicators. The fitness value is generated through weighted summation. The weight coefficient is dynamically adjusted according to the system operation stage (for example, transmission bandwidth is given a higher weight in the real-time priority stage).

[0046] Collaborative optimization process: Initialize the population: randomly generate multiple sets of parameter combinations as initial chromosomes; Crossover and mutation: Perform arithmetic crossover and Gaussian mutation operations on high fitness chromosomes; Parameter output: The optimal chromosome after iteration is decoded into a collaborative parameter set and sent to each module synchronously.

[0047] Technical Logic: By jointly encoding multi-module parameters into chromosomes, this approach solves the system-level performance imbalances caused by traditional single-module optimization. For example, when the dynamics of a coal pile intensify, the genetic algorithm automatically increases the scanning frequency and reduces modeling accuracy, while simultaneously increasing the compression ratio to balance real-time performance and accuracy requirements, achieving global optimization without manual intervention.

[0048] Feature selection unit: Gene pool construction: Geometric features (curvature, normal vector), temporal features (point cloud displacement rate), and environmental features (dust concentration gradient) are defined as candidate feature sets, and each feature is assigned an independent gene fragment and stored in the gene pool.

[0049] Feature combination screening: Using the roulette wheel selection algorithm, the selection probability is assigned based on the contribution of features in historical data to modeling accuracy, generating multiple sets of candidate feature combinations.

[0050] Weight matrix generation: Normalize the selected features and construct a weight matrix based on inter-feature correlation analysis (such as the Pearson coefficient) to guide the layered transmission module to adopt differentiated compression strategies for different features.

[0051] The probabilistic screening mechanism of the genetic algorithm addresses the subjectivity and inefficiency of manual feature engineering. For example, in scenarios where dust concentration suddenly increases, the algorithm automatically prioritizes environmental features, prioritizing the transmission of dust distribution data to support analysis by the anomaly detection module while deprioritizing the transmission of geometric features, thereby maximizing data value within limited bandwidth.

[0052] Path planning unit: Deformation area identification: Based on the curvature change rate of the coal pile surface output by the real-time modeling module, the deformation area bounding box is extracted through threshold segmentation.

[0053] Adaptive path generation: A greedy algorithm is used to generate dense scanning paths (such as spiral trajectories) in the deformable area, sparse raster scanning is used in the stable area, and the gimbal steering angle and movement speed are dynamically calculated.

[0054] Trajectory feedback mechanism: Feedback the path execution results (such as the coordinates of uncovered areas) to the genetic algorithm as the basis for fitness evaluation in the next round of path optimization.

[0055] The closed loop of "deformation detection-path generation-execution feedback" eliminates the resource waste associated with fixed scanning paths. For example, when a collapsed edge of a coal pile is detected, the algorithm automatically increases the scanning point density in that area while reducing repeated scanning of flat areas, improving data acquisition efficiency by over 30%. (Note: This is only an example; actual implementation does not require quantitative data.)

[0056] Model evolution unit: Neural Network Architecture Search (NAS): Defines the search space including hyperparameters such as the number of convolutional layers, pooling method, and number of fully connected nodes. Generates and evaluates the detection accuracy of different network structures on historical anomaly datasets through genetic algorithms.

[0057] Model parameter migration: Encapsulate the optimized network structure (such as three-layer convolution + maximum pooling + 128-node full connection) and its weight parameters into a lightweight inference model and deploy it to the anomaly detection module.

[0058] Online update mechanism: Regularly collect the latest false positive / missed negative samples and update model parameters through fine-tuning to prevent model aging.

[0059] Through automated architecture search and continuous learning, the generalization issues of traditional anomaly detection models are addressed. For example, a genetic algorithm can rapidly evolve an adaptive network structure for a new anomaly pattern, such as slow surface subsidence on a coal pile. This reduces the false positive rate by approximately 25% compared to models with fixed structures (note: these values are illustrative only).

[0060] Global drive of genetic algorithm: through the parameter optimization unit to coordinate the operation strategies of multiple modules, the feature selection unit to optimize the data transmission content, the path planning unit to improve the efficiency of data collection, and the model evolution unit to enhance the anomaly detection capability, a closed-loop optimization system is formed.

[0061] Dynamic adaptability: Each unit dynamically adjusts its strategy based on real-time data and feedback. For example, the feature weight matrix is updated as the environment changes, and the scanning path is dynamically optimized as the coal pile deforms.

[0062] Module decoupling and collaboration: Each unit independently implements specific functions (such as parameter encoding and feature screening), while cross-unit collaboration is achieved through chromosome interaction and fitness sharing of genetic algorithms.

[0063] Based on the above implementations, those skilled in the art can implement specific systems by combining conventional genetic algorithm libraries (such as DEAP), point cloud processing frameworks (such as PCL), and deep learning frameworks (such as TensorFlow Lite), and complete module integration and parameter tuning without creative work.

[0064] Specifically, the intelligent laser coal processing system of the present invention includes a hierarchical data transmission module comprising a cascaded data cleaning unit, a dynamic compression unit, and an adaptive sharding unit: The data cleaning unit receives the raw data stream from the data acquisition module and removes outliers based on the preset dust concentration threshold and statistical filtering algorithm; The dynamic compression unit adopts a layered compression strategy for the cleaned data and optimizes the space division and encoding parameters based on the genetic algorithm; The adaptive sharding unit dynamically adjusts the data sharding granularity and adds forward error correction code based on the network delay prediction value fed back by the real-time modeling module.

[0065] Data cleaning unit: Dust interference filtration mechanism: Threshold determination: Based on the preset dust concentration threshold (e.g. ≥50mg / m³), the contaminated data segment is marked and the outlier removal process is triggered; Statistical filtering algorithm: Use density-based clustering algorithms (such as DBSCAN) to identify isolated noise points in point cloud data, and combine time window sliding average method to smooth outliers in time series; Data calibration: Perform sensor calibration compensation on the retained valid data segments to correct the laser ranging deviation caused by dust refraction.

[0066] Abnormal recovery mechanism: When N consecutive frames of data are judged to be invalid, redundant sensor data interpolation filling is started.

[0067] The dual mechanism of "threshold determination + cluster filtering" solves the problem of data distortion in high-dust environments. For example, when a sudden increase in dust concentration causes discrete noise points to appear in the lidar point cloud, the DBSCAN algorithm can effectively identify and remove outliers far from the main point cloud cluster, while also avoiding data interruptions through interpolation compensation, providing high-quality input for subsequent processing.

[0068] Dynamic Compression Unit: Tiered compression strategy: The first layer of spatial division: KD-Tree is used to spatially partition the point cloud data and merge adjacent point cloud clusters; The second layer of feature dimensionality reduction: Principal component analysis (PCA) is applied to each partition, retaining the main direction eigenvector; The third layer of hybrid coding: the data after dimensionality reduction is compressed using a combination of run-length encoding (RLE) and Huffman coding.

[0069] Genetic algorithm optimization: Chromosome encoding: Map the KD-Tree depth, PCA retained dimensions, and encoding method selection to chromosome gene positions; Fitness function: Construct a multi-objective evaluation function based on compression rate, data reconstruction error, and computation time; Parameter generation: Through multiple generations of evolution, the optimal parameter combination that adapts to the current network status and modeling requirements is selected.

[0070] The three-stage compression method of "spatial partitioning, feature dimensionality reduction, and hybrid encoding" addresses the bandwidth bottleneck in transmitting massive point cloud data. A genetic algorithm dynamically optimizes compression parameters, for example automatically increasing the KD-Tree partition depth in complex areas around the coal pile's edge to preserve detailed features, while reducing the partition depth in flat areas to improve compression, achieving an intelligent balance between fidelity and efficiency.

[0071] Adaptive Sharding Unit: Sharding granularity adjustment strategy: Preamble design: Add metadata including the shard size and error correction code type to the header of each data shard; Dynamic sharding rules: Based on the network delay prediction value fed back by the real-time modeling module, a table lookup method is used to determine the shard size (for example, when the delay is ≤ 50ms, the shard size is 1KB; when the delay is > 100ms, the shard size is 512B); Error correction code addition: Forward error correction codes (such as Reed-Solomon codes) are used to add redundant check bits to the sharded data.

[0072] Feedback mechanism: The fragment transmission success rate and error correction code decoding failure rate are fed back to the genetic algorithm driving engine as auxiliary indicators for compression parameter optimization.

[0073] The "dynamic sharding + error correction code" mechanism solves the problem of packet loss caused by network fluctuations. For example, when network latency increases, the sharding granularity is automatically reduced and the error correction code redundancy is increased. Even if some shards are lost, the complete data can still be restored through the error correction code, ensuring that the modeling module receives a continuous data stream.

[0074] Overall logical summary of the technical solution Module cascade processing flow: Data cleaning: ensure the validity and accuracy of input data; Dynamic compression: improves data transmission efficiency and reduces bandwidth usage; Adaptive Sharding: Enhances data transmission robustness to adapt to network fluctuations.

[0075] Genetic algorithm collaborative optimization: The compression parameters (KD-Tree depth, PCA dimension, etc.) and sharding strategy (shard size, error correction strength) are jointly optimized through chromosome encoding to form a global optimal solution.

[0076] Closed-loop feedback mechanism: The transmission success rate of the sharding unit is fed back to the compression unit to guide the direction of the next round of parameter evolution; The network delay prediction value of the modeling module dynamically adjusts the sharding strategy to form an adaptive closed loop.

[0077] Implementation effect example: In a scenario where the surface of a coal pile changes dynamically, the system uses a dynamic compression unit to reduce the amount of data in flat areas (the compression ratio is increased to 60%). At the same time, it uses an adaptive sharding unit to maintain a transmission success rate of over 95% when the delay fluctuates, ultimately ensuring that the modeling module obtains complete and high-precision input data.

[0078] Specifically, the intelligent laser coal processing system of the present invention includes the following real-time modeling and analysis module: The difference detection unit locates the changed area reported by the data acquisition module based on the spatial hash table optimized by the genetic algorithm; The multi-granularity modeling unit generates a multi-resolution coal pile model by synchronously optimizing octree voxelization and Poisson surface reconstruction parameters through dynamic chromosome encoding; The trend prediction unit uses a non-dominated sorting genetic algorithm to balance modeling error and computing resource consumption, and outputs the prediction results to the anomaly detection module.

[0079] Difference detection unit: Spatial hash table construction: 3D grid division: The coal pile scanning area is divided into spatial grids according to a preset benchmark resolution (e.g., 0.5m×0.5m×0.5m), and each grid is assigned a unique hash value; Dynamic update mechanism: Update the point cloud density statistics (such as number of points and average height) within the grid based on the point cloud data reported by the data acquisition module; Change area location: Compare the difference between the current grid density and the historical benchmark model. If the difference exceeds the threshold, it is marked as a change area.

[0080] Genetic algorithm optimization: Chromosome encoding: mapping the hash table grid size, difference threshold, and density statistics window length to chromosome gene bits; Fitness function: Construct a multi-objective optimization function based on the detection accuracy (missed detection rate / false detection rate) of the changed area and the computational time; Parameter iteration: The optimal combination of grid division and difference judgment parameters is screened out through crossover and mutation operations.

[0081] Technical Logic: A spatial hash table is used to quickly locate dynamically changing areas on the coal pile surface, addressing the efficiency issues associated with full data comparison. For example, when new coal is unloaded from a localized area of the coal pile, the hash table only compares data against the changed mesh, reducing computational time by approximately 70% (Note: These values are examples only and do not require quantification). A genetic algorithm dynamically optimizes meshing parameters to avoid missing small deformations or over-detecting flat areas due to fixed resolution.

[0082] Multi-granularity modeling unit: Octree voxelization process: Coarse-grained segmentation: The entire space of the coal pile is recursively segmented using an octree, with an initial voxel size of 1m³; Dynamic refinement: Sub-octree segmentation is performed on the changed areas marked by the difference detection unit, and the minimum voxel size can be refined to 0.1m³; Feature extraction: Count the curvature and normal vector distribution characteristics of the point cloud within each voxel.

[0083] Poisson surface reconstruction optimization: Point cloud sampling: Dynamically adjust the sampling point spacing of Poisson reconstruction according to the voxel feature distribution density; Surface Smoothing: Optimize surface continuity and smoothness through normal vector consistency constraints.

[0084] Dynamic chromosome encoding: The voxel refinement depth, sampling spacing, and smoothing coefficient are encoded into chromosomes, and parameter combinations are iteratively generated through a genetic algorithm; Modeling accuracy (surface fitting residual) and computing resource consumption (memory usage, CPU utilization) are used as fitness evaluation indicators.

[0085] This hierarchical modeling strategy of "global coarse-grained + local fine-grained" addresses the difficulty traditional methods face in balancing accuracy and efficiency in complex coal pile configurations. For example, in collapsed areas at the edge of a coal pile, voxel resolution is automatically increased to capture detailed deformation, while maintaining low resolution in stable areas, improving overall modeling efficiency by approximately 40%. A genetic algorithm optimizes parameter combinations, ensuring automatic adaptation of the optimal modeling strategy for different scenarios.

[0086] Trend Forecasting Unit: Non-dominated Sorting Genetic Algorithm (NSGA-II) Applications: Objective function definition: Objective 1: Modeling error (RMS deviation between the current model and historical data); Objective 2: Computational resource consumption (CPU / GPU utilization, memory usage); Chromosome encoding: Encode modeling frequency, maximum voxel refinement depth, and surface smoothing weight into multi-objective optimization chromosomes; Pareto front screening: Select the optimal solution set through non-dominated sorting and congestion calculation to generate multiple sets of feasible parameter combinations.

[0087] Prediction model output: Predict future modeling resource requirements based on the optimal parameter combination and dynamically adjust computing node allocation (such as GPU acceleration resource scheduling); Output the deformation trend heat map to the anomaly detection module to mark high-risk areas where landslides or accumulation overloads may occur.

[0088] Technical Logic: By balancing modeling accuracy and resource consumption through multi-objective optimization, NSGA-II overcomes the single-goal-oriented limitations of traditional forecasting models. For example, when system computing resources are limited, NSGA-II automatically selects a "medium accuracy + low resource usage" parameter combination to ensure the real-time responsiveness of the anomaly detection module. The generation of trend heat maps enables operations and maintenance personnel to proactively intervene in high-risk areas and avoid safety incidents.

[0089] Overall logical summary of the technical solution: Module collaboration process: Difference detection: quickly locate the changed area and narrow the scope of modeling calculation; Multi-granularity modeling: Dynamically adjust modeling accuracy based on changing regions to generate multi-resolution models; Trend prediction: Optimize resource allocation strategies and guide the system to dynamically adapt operating parameters.

[0090] Genetic algorithm runs through optimization: The difference detection unit optimizes hash table parameters; The multi-granularity modeling unit simultaneously optimizes voxelization and surface reconstruction parameters; The trend prediction unit realizes multi-objective resource allocation decision-making.

[0091] Real-time and precision assurance: Through hierarchical modeling and dynamic parameter adjustment, modeling efficiency is maximized under limited resources while ensuring high-precision expression of key areas.

[0092] Example of implementation effect: When continuous unloading operations occur in a coal pile, the difference detection unit locates the unloading area → the multi-granularity modeling unit performs high-precision voxel segmentation and surface reconstruction on the area → the trend prediction unit predicts the unloading rate and adjusts the modeling frequency, and finally outputs a real-time updated three-dimensional model and deformation warning signal.

[0093] Specifically, in the intelligent laser coal processing system of the present invention, the closed-loop optimization module is implemented through a feedback loop: The federated evolution unit collects anomaly detection data from each coal yard node and performs chromosome migration through a differential privacy protocol; Dynamic weight allocation unit, which dynamically adjusts the objective function weight of the detection model according to the false positive or false negative statistics of the early warning module; The feedback learning unit injects the optimized parameter set back into the genetic algorithm driving engine to form a closed loop of parameter evolution.

[0094] Federation Evolution Unit: Cross-node data collection: Each coal yard node periodically uploads anomaly detection data (such as deformation area coordinates and false alarm type labels). The data format is standardized into a JSON structure and timestamps are added. The central server aggregates multi-node data through lightweight communication protocols (such as MQTT) to generate a global abnormal feature distribution map.

[0095] Differential privacy protection mechanism: Chromosome desensitization: Add Laplace noise to the chromosome segments to be migrated (such as feature selection weights and modeling parameters) to meet the ε-differential privacy constraint; Secure transmission: Chromosome data is transmitted through a TLS encrypted channel to prevent intermediate nodes from stealing sensitive information.

[0096] Co-evolution process: The central server generates candidate populations based on multi-node chromosomes and integrates optimization features of different coal fields through crossover operations; The evolved chromosomes are distributed to each node to replace the low-fitness individuals in the local chromosome pool.

[0097] Through a "data desensitization-co-evolution-parameter distribution" mechanism, the conflict between data privacy and scenario adaptation during the collaborative optimization of multiple coal yards is resolved. For example, when a coal yard evolves highly robust parameters due to unique geological conditions, differential privacy technology can migrate these parameter features to other nodes without exposing the original data, improving the overall adaptability of the system.

[0098] Dynamic weight allocation unit: False positive / missing negative statistics: False alarm judgment: After the warning is triggered, no actual deformation or equipment abnormality is detected within the preset time; Missed alarm determination: No warning was triggered but subsequent manual inspection confirmed the presence of an abnormality; The statistical window is set to a sliding time window (such as the last 24 hours) and the false positive rate (FPR) and false negative rate (FNR) are calculated.

[0099] Dynamic weight adjustment: Objective function reconstruction: The loss function of the detection model is defined as a weighted sum: Loss = α×FPR + β×FNR + γ×computation delay; Adaptive weighting: Automatically increase the α weight when the FPR increases, and increase the β weight when the FNR increases, and iteratively optimize the coefficients through gradient descent method; Weight limit: Set upper and lower limits for weights (such as α∈[0.3,0.7]) to prevent a single indicator from excessively affecting model stability.

[0100] By dynamically balancing the penalty weights for false positives and false negatives, the system addresses the detection bias inherent in traditional fixed-weight models when operating conditions change. For example, during the rainy season, when large fluctuations in coal pile humidity lead to an increase in false positives, the system automatically reduces the α weight to reduce excessive warnings while maintaining false negative monitoring sensitivity.

[0101] Feedback Learning Unit: Parameter back injection: Parameter packaging: Package the optimized weight coefficients, chromosome segments, and model hyperparameters into structured data packets; Version control: Add a version identifier for each parameter update to support rapid rollback in abnormal conditions; Priority scheduling: Inject parameters in different levels according to their impact range (e.g. global parameters are updated first, while local parameters are updated later).

[0102] Closed-loop update mechanism: Every time a genetic algorithm iteration cycle is completed (e.g., 8 hours), a parameter synchronization instruction is triggered; Broadcast the parameter package to the genetic algorithm driving engine and various functional modules through a message queue (such as Kafka); After the receiving module verifies the data integrity, it replaces the runtime parameters in the memory.

[0103] Standardized parameter management and reliable transmission mechanisms ensure that optimization results are effectively implemented. For example, when a new chromosome parameter set significantly reduces modeling error, the system completes the full node parameter update within the preset maintenance window, avoiding instant service interruptions caused by online updates.

[0104] Overall logical summary of the technical solution: Module collaboration process: Federated Evolution: Achieve cross-site knowledge sharing and privacy protection; Weight allocation: Dynamically optimize the goal-oriented detection model; Feedback learning: Ensures that optimization strategies are implemented synchronously across the entire system.

[0105] Key technical features: Privacy-performance balance: Achieving a balance between data availability and privacy protection through differential privacy; Adaptability: Dynamically adjust system behavior based on real-time statistical indicators; Robustness guarantee: Version control and priority scheduling prevent system shocks caused by parameter updates.

[0106] Implementation effect example: When the missed alarm rate in a coal yard increases due to aging equipment, the dynamic weight allocation unit automatically increases the missed alarm penalty weight → the feedback learning unit injects the new weight into the detection module → the federated evolution unit migrates the optimization strategy to other nodes through desensitization processing, ultimately achieving a coordinated improvement in the detection accuracy of the entire system.

[0107] In a second aspect, the present invention provides an intelligent laser coal processing method, which is applied to the intelligent laser coal processing system, comprising: S1: Real-time environmental data is acquired through the distributed sensors of the data acquisition module to trigger redundant equipment switching and dynamic compensation; S2: Input the data collected by S1 into the hierarchical transmission module and use genetic algorithm to optimize feature selection and data compression strategy; S3: Based on the feature data processed by S2, a real-time 3D model is generated through incremental modeling and multi-granularity reconstruction in the modeling and analysis module; S4: Input the modeling results of S3 into the anomaly detection module, combine the knowledge graph and evolutionary model to identify abnormal features and trigger an alert; S5: Based on the early warning feedback from S4, the closed-loop optimization module dynamically adjusts the system parameters and updates the genetic algorithm evolution strategy.

[0108] Specifically, the intelligent laser coal processing method of the present invention, S2 specifically includes: S21: Receives the dust concentration and gimbal motion parameter data transmitted by S1, and screens the optimal combination of geometric features and environmental features through the gene pool of the genetic algorithm; S22: Use a hierarchical compression strategy to perform KD-tree partitioning, PCA dimensionality reduction, and hybrid encoding on the selected features; S23: Dynamically adjust the compression ratio and sharding granularity based on the computing resource status fed back by the S3 module.

[0109] Specifically, the intelligent laser coal processing method of the present invention, S3 includes: S31: Build a spatial hash table based on the pre-processed data in S2 to locate the deformation area reported by the data acquisition module; S32: Simultaneous optimization of voxelization resolution and surface reconstruction parameters via dynamic chromosome encoding; S33: Use the NSGA-II algorithm to balance modeling accuracy and resource consumption, and output the optimized model to S4.

[0110] Specifically, in the intelligent laser coal processing method of the present invention, the step S4 executes: S41: Establish a Bayesian network association model of equipment status, environmental parameters and modeling errors; S42: Monte Carlo tree search for abnormal root cause analysis based on the deformation prediction results of S3; S43: Update the detection model parameters using the false positive data fed back from S5.

[0111] Specifically, the intelligent laser coal processing method of the present invention, S5 includes: S51: Collect performance data of each module every 24 hours and execute genetic algorithm evolution iteration; S52: Using differential privacy protocol to achieve cross-coal yard model parameter migration; S53: Use the false positive and negative data generated by S4 as the basis for fitness evaluation and reversely optimize the feature selection strategy.

[0112] S1: Data acquisition and dynamic compensation: Sensor collaborative collection: Distributed laser radar and infrared sensors synchronously collect point cloud data, and the pan-tilt encoder provides real-time feedback on rotation angle and movement speed; The dust concentration sensor samples at a frequency of 1 Hz and triggers the threshold judgment logic.

[0113] Redundancy switching and compensation: When the dust concentration exceeds the threshold, the system switches to the backup lens and activates the pan / tilt angle compensation to ensure that the scanning covers no blind spots. Dynamically adjust the lidar scanning frequency (for example, lower the frequency to reduce noise when dust concentration is high).

[0114] Multi-sensor collaboration and redundancy ensure continuous data collection under extreme working conditions. For example, if the primary lens fails due to dust, the system seamlessly switches to the backup lens and maintains the integrity of the scanning trajectory through pan-tilt angle compensation, providing a reliable data source for subsequent processing.

[0115] S2: Layered data transmission and optimization: Gene pool screening: Construct a gene pool containing geometric features (curvature, normal vector) and environmental features (dust distribution gradient); The roulette wheel selection strategy based on genetic algorithm is used to screen out feature combinations that have high contribution to the current modeling task.

[0116] Tiered compression strategy: Spatial partitioning: KD-Tree is used to divide the point cloud into sub-regions and merge low feature complexity regions; Dimensionality reduction coding: PCA is applied to each sub-region to extract the principal component features, and run-length encoding (RLE) is combined to compress the data volume; Dynamic Adjustment: Automatically select high compression ratio or high fidelity mode based on GPU utilization feedback from the modeling module.

[0117] Technical Logic: Data transmission efficiency is optimized through a three-level process: feature screening, spatial compression, and dynamic adaptation. For example, high-precision geometric features are retained in complex areas around the edge of a coal pile, while a high compression ratio strategy is used in flat areas to reduce bandwidth usage while meeting modeling requirements.

[0118] S3: Incremental Modeling and Multi-Grained Reconstruction: Spatial hash table construction: The coal pile space is divided into reference grids (e.g., 0.5 m³), and the hash table records the point cloud density and height mean of each grid; Comparing the real-time data with the benchmark model, the grids with location density changes exceeding 10% are considered deformed areas.

[0119] Parameter synchronization optimization: Encode voxel resolution (e.g. 0.1m-1m) and surface smoothness coefficients into dynamic chromosomes; An elite retention strategy is used to iteratively screen the Pareto optimal solution set, balancing modeling accuracy and computational overhead.

[0120] Technical Logic: Efficient modeling is achieved through "local updates + parameter optimization." For example, high-resolution voxelization is performed only on the deformed area, while the remaining areas use the historical model. This reduces computational effort by 70% while maintaining accuracy (the value is for example only).

[0121] S4: Anomaly Detection and Root Cause Analysis: Bayesian network construction: Define node variables: device status (gimbal current, motor temperature), environmental parameters (temperature, humidity, dust concentration), modeling error (surface residual); The conditional probability table is trained based on historical data to establish multi-factor association relationships.

[0122] Monte Carlo Tree Search (MCTS): Simulate deformation propagation paths and calculate the probability of abnormalities caused by different equipment failures or environmental factors; Output root cause hypotheses (e.g., “gimbal jam resulting in scan loss” or “dust interference causing modeling errors”).

[0123] Improve the accuracy of anomaly diagnosis through knowledge reasoning and simulation search. For example, when a sudden increase in local modeling error is detected, the Bayesian network, combined with abnormal PTZ current data, infers mechanical failure as the root cause and triggers targeted maintenance instructions.

[0124] S5: Closed-loop optimization and parameter evolution: Differential Privacy Migration: Add Laplace noise (privacy budget ε=0.5) to the model parameters to be migrated, desensitize them, and upload them to the central server; The server aggregates multi-node parameters and generates an enhanced chromosome library through crossover operations.

[0125] Feedback optimization: Add false positive / missing negative data to the genetic algorithm fitness function to reduce the weight of similar feature combinations; Global parameter updates are triggered every 24 hours and broadcast to each subsystem through the message queue.

[0126] Technical Logic: Achieve cross-site collaborative evolution and self-learning. For example, an optimized feature selection strategy at a coal yard is migrated to other nodes with privacy protection, significantly improving the system's sensitivity to detecting new abnormal patterns.

[0127] Overall logical summary of the technical solution: Full process closed loop: Data is collected, transmitted, modeled, tested, and optimized in a closed loop, with each step dynamically adjusting strategies through a feedback mechanism. Genetic algorithms are used throughout to achieve global parameter optimization and cross-scenario adaptation.

[0128] Core innovations: Layered compression and dynamic adjustment: Resolve the contradiction between data transmission bandwidth and modeling accuracy; Incremental modeling and multi-granularity optimization: balancing computing resources and model accuracy requirements; Federated evolution and privacy protection: achieving cross-site knowledge sharing and data security collaboration.

[0129] Implementation effect: The system can automatically adapt to the high-dust and changeable coal yard environment, reducing manual intervention; Improve anomaly detection accuracy and warning timeliness through continuous parameter evolution.

[0130] The present invention effectively solves the problems of sensor failure and motor jamming under high dust conditions; it utilizes the genetic algorithm of the hierarchical data transmission module to optimize the feature compression strategy and adaptive sharding technology, combined with the incremental spatial hash table positioning and multi-granularity parameter synchronization optimization of the real-time modeling and analysis module, to break through the delay bottleneck of traditional full-scale modeling; the anomaly detection module associates modeling errors with environmental parameters through the Bayesian network, combined with Monte Carlo tree search root cause analysis to improve the accuracy of early warning; the closed-loop optimization module adopts the federated evolution mechanism to realize cross-station parameter migration, and continuously optimizes the system parameters through dynamic weight allocation and feedback learning, and finally realizes millisecond-level response of coal pile deformation monitoring under extreme working conditions, synchronously guarantees modeling accuracy and system robustness, and completely solves the technical pain points of poor real-time performance, high false alarm and missed alarm rates, and weak environmental adaptability of traditional systems.

Claims

1. An intelligent laser coal processing system, characterized in that: It includes an environment-adaptive data acquisition module, a hierarchical data transmission module, a real-time modeling and analysis module, an anomaly detection and early warning module, and a closed-loop optimization module and a genetic algorithm driving engine that interact with each module in two-way data. The environmental adaptive data acquisition module collects dust concentration and pan / tilt motion parameters in real time through a distributed sensor network, triggers redundant lens switching and dynamic compensation according to preset thresholds, and outputs the original data stream to the layered data transmission module; After receiving the original data stream, the layered data transmission module generates standardized feature data through edge preprocessing, optimizes feature selection and compression strategies based on genetic algorithms, and transmits the compressed valid data to the real-time modeling and analysis module; The real-time modeling and analysis module constructs a reference three-dimensional grid based on an incremental modeling engine, combines the characteristic data provided by the hierarchical data transmission module, and uses a genetic algorithm to optimize local reconstruction parameters to generate a real-time coal pile model; The anomaly detection and warning module receives the real-time coal pile model and sensor data, associates multi-source data features through the knowledge graph, and triggers a hierarchical warning signal to be fed back to the closed-loop optimization module; The closed-loop optimization module works in conjunction with the genetic algorithm driving engine to dynamically optimize the parameters of each module based on early warning signals and system performance indicators.

2. The intelligent laser coal processing system according to claim 1 is characterized in that: The genetic algorithm driven engine includes: The parameter optimization unit, based on the chromosome encoding mechanism of the genetic algorithm, jointly optimizes the scanning frequency of the data acquisition module, the accuracy level of the modeling module, and the compression ratio of the transmission module to generate a collaborative parameter set; The feature selection unit screens the optimal combination of geometric features, temporal features, and environmental features based on the gene pool to generate a feature compression weight matrix for the hierarchical transmission module; The path planning unit generates an adaptive scanning path based on the deformation detection results of the real-time modeling module and dynamically adjusts the pan-tilt motion trajectory of the data acquisition module; The model evolution unit uses the neural network architecture to search and optimize the anomaly detection model, and feeds the evolved model parameters back to the early warning module.

3. The intelligent laser coal processing system according to claim 1 is characterized in that: The layered data transmission module includes a data cleaning unit, a dynamic compression unit and an adaptive fragmentation unit for cascade processing: The data cleaning unit receives the raw data stream from the data acquisition module and removes outliers based on the preset dust concentration threshold and statistical filtering algorithm; The dynamic compression unit adopts a layered compression strategy for the cleaned data and optimizes the space division and encoding parameters based on the genetic algorithm; The adaptive sharding unit dynamically adjusts the data sharding granularity and adds forward error correction code based on the network delay prediction value fed back by the real-time modeling module.

4. The intelligent laser coal processing system according to claim 1 is characterized in that: The real-time modeling and analysis module includes iterative optimization: The difference detection unit locates the changed area reported by the data acquisition module based on the spatial hash table optimized by the genetic algorithm; The multi-granularity modeling unit generates a multi-resolution coal pile model by synchronously optimizing octree voxelization and Poisson surface reconstruction parameters through dynamic chromosome encoding; The trend prediction unit uses a non-dominated sorting genetic algorithm to balance modeling error and computing resource consumption, and outputs the prediction results to the anomaly detection module.

5. The intelligent laser coal processing system according to claim 1 is characterized in that: The closed-loop optimization module is implemented through a feedback loop: The federated evolution unit collects anomaly detection data from each coal yard node and performs chromosome migration through a differential privacy protocol; Dynamic weight allocation unit, which dynamically adjusts the objective function weight of the detection model according to the false positive or false negative statistics of the early warning module; The feedback learning unit injects the optimized parameter set back into the genetic algorithm driving engine to form a closed loop of parameter evolution.

6. An intelligent laser coal processing method, applied to the intelligent laser coal processing system according to any one of claims 1 to 5, characterized in that: include: S1: Real-time environmental data is acquired through the distributed sensors of the data acquisition module to trigger redundant equipment switching and dynamic compensation; S2: Input the data collected by S1 into the hierarchical transmission module and use genetic algorithm to optimize feature selection and data compression strategy; S3: Based on the feature data processed by S2, a real-time 3D model is generated through incremental modeling and multi-granularity reconstruction in the modeling and analysis module; S4: Input the modeling results of S3 into the anomaly detection module, combine the knowledge graph and evolutionary model to identify abnormal features and trigger an alert; S5: Based on the early warning feedback from S4, the closed-loop optimization module dynamically adjusts the system parameters and updates the genetic algorithm evolution strategy.

7. The intelligent laser coal processing method according to claim 6 is characterized in that: The S2 specifically includes: S21: Receives the dust concentration and gimbal motion parameter data transmitted by S1, and screens the optimal combination of geometric features and environmental features through the gene pool of the genetic algorithm; S22: Use a hierarchical compression strategy to perform KD-tree partitioning, PCA dimensionality reduction, and hybrid encoding on the selected features; S23: Dynamically adjust the compression ratio and sharding granularity based on the computing resource status fed back by the S3 module.

8. The intelligent laser coal processing method according to claim 6 is characterized in that: The S3 includes: S31: Build a spatial hash table based on the pre-processed data in S2 to locate the deformation area reported by the data acquisition module; S32: Simultaneous optimization of voxelization resolution and surface reconstruction parameters via dynamic chromosome encoding; S33: Use the NSGA-II algorithm to balance modeling accuracy and resource consumption, and output the optimized model to S4.

9. The intelligent laser coal processing method according to claim 6, characterized in that: The S4 performs: S41: Establish a Bayesian network association model of equipment status, environmental parameters and modeling errors; S42: Monte Carlo tree search for abnormal root cause analysis based on the deformation prediction results of S3; S43: Update the detection model parameters using the false positive data fed back from S5.

10. The intelligent laser coal processing method according to claim 6, characterized in that: The S5 comprises: S51: Collect performance data of each module every 24 hours and execute genetic algorithm evolution iteration; S52: Using differential privacy protocol to achieve cross-coal yard model parameter migration; S53: Use the false positive and negative data generated by S4 as the basis for fitness evaluation and reversely optimize the feature selection strategy.

Images