Blasting data multi-dimensional analysis and anomaly detection processing system

Abstract

The invention discloses a blasting data multi-dimensional analysis and anomaly detection processing system, and particularly relates to the technical field of blasting intelligent analysis processing, and the system comprises a distributed data collection module, a data preprocessing module, a multi-dimensional analysis engine, a hybrid anomaly detection module, a visual interaction interface, an early warning processing module, and an extensible interface module. Aiming at insufficient multi-source heterogeneous data fusion capability in the prior art, the invention innovatively adopts a WPT-EMD combined denoising algorithm and a dynamic standardization unit to work cooperatively, and effectively overcomes the problems of mode aliasing and noise residual in the traditional method through fine frequency band division of wavelet packet transform (WPT) and self-adaptive characteristics of empirical mode decomposition (EMD); in combination with a dynamic standardization strategy based on information entropy, feature space alignment of heterogeneous data such as vibration waveforms and stress fields is realized, the signal-to-noise ratio of original data is improved, the feature dimension consistency is improved, and a high-quality data foundation is laid for subsequent analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of blasting intelligent analysis and processing, and more specifically, to a blasting data multi-dimensional analysis and anomaly detection processing system. Background Art

[0002] The multi-dimensional analysis and anomaly detection and processing system for blasting data is of great significance. By real-time monitoring of multi-dimensional data such as explosives usage, detonation time, and vibration intensity, it can promptly warn of abnormal situations, effectively prevent safety accidents, and protect the safety of life and property. At the same time, the system can optimize blasting parameters, control operation quality, provide data support for management, and achieve scientific decision-making and cost control. In addition, it can ensure that blasting operations comply with regulatory standards and reduce environmental impact. The data stored in the system is conducive to knowledge inheritance and technological innovation, contributing to the sustainable development of the blasting industry.

[0003] The existing technology has the following defects:

[0004] 1. Insufficient multi-source heterogeneous data fusion capabilities: Traditional systems typically analyze data from a single sensor type and lack a collaborative processing mechanism for multimodal data such as vibration waveforms, stress field distributions, and geological radar data. Existing data preprocessing methods are mostly based on fixed threshold denoising and linear normalization, which are difficult to effectively handle non-stationary noise interference and dimensional differences between heterogeneous data sources, resulting in severe information loss during feature extraction.

[0005] 2. Limited generalization capabilities of anomaly detection models: Existing methods often rely on single algorithms (such as isolation forests or support vector machines), which are difficult to handle with the strong spatiotemporal correlations and dynamic changes in anomaly patterns in explosive data. Traditional isolation forests are prone to misjudgment in high-dimensional feature spaces, while deep learning methods based on reconstruction errors are inadequate in modeling temporal dependencies and lack multi-dimensional evidence fusion mechanisms.

[0006] 3. Weak support for human-machine collaborative decision-making: Existing system visualization is mostly limited to two-dimensional charts, lacking intuitive representation of the three-dimensional dynamic evolution of the blasting process and the propagation path of anomalies. Traditional early warning modules typically use fixed threshold alarms and fail to incorporate contextual information such as geological conditions and environmental parameters for risk assessment. Furthermore, the generation of response recommendations relies on manual experience and lacks intelligent matching of case-based reasoning with real-time working conditions.

[0007] 4. Lack of system scalability and self-optimization capabilities: Traditional architectures often use a tightly coupled design, and algorithm updates require downtime for maintenance, making it difficult to support the rapid integration of new sensors and detection models. Existing learning mechanisms mostly rely on offline batch training, which cannot achieve dynamic adjustment of model parameters, resulting in degraded detection performance after long-term system operation.

[0008] Therefore, to address the above problems, a blasting data multi-dimensional analysis and anomaly detection processing system is proposed. Summary of the Invention

[0009] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a blasting data multi-dimensional analysis and anomaly detection processing system to solve the problems raised in the above-mentioned background technology.

[0010] To achieve the above-mentioned objectives, the present invention provides the following technical solutions: a multi-dimensional analysis and anomaly detection processing system for blasting data, comprising: a distributed data acquisition module, a data preprocessing module, a multi-dimensional analysis engine, a hybrid anomaly detection module, a visual interactive interface, an early warning and disposal module, and an extensible interface module; the distributed data acquisition module is connected to a vibration sensor array, a geological radar, and an unmanned aerial survey device through a multi-protocol adapter; the data preprocessing module establishes a unified spatiotemporal benchmark for the collected raw data; the multi-dimensional analysis engine maps the processed data into a feature space to construct a dynamic knowledge graph; the hybrid anomaly detection module implements cross-dimensional reasoning based on the node relationship of the knowledge graph; the visual interactive interface and the self-learning optimization module form a human-machine collaborative decision-making loop; the early warning and disposal module connects to the emergency control system to form a closed disposal loop; and the extensible interface module provides standardized expansion channels for each component.

[0011] Preferably, the data preprocessing module adopts a wavelet packet transform-empirical mode decomposition (WPT-EMD) joint denoising algorithm, processes non-stationary noise by constructing an adaptive threshold function, and is provided with a dynamic normalization unit based on feature importance. The unit calculates the information entropy weight of each measuring point through a sliding time window, performs nonlinear normalization on the vibration waveform data, and adopts distribution alignment based on kernel density estimation for discrete stress monitoring data to achieve feature space consistency conversion of heterogeneous data.

[0012] Preferably, the multidimensional analysis engine is implemented through multimodal feature fusion technology: time domain feature extraction adopts the improved HHT transform that integrates the Marginal Spectrum entropy value, spatial domain analysis adopts the Delaunay triangulation algorithm combined with Kriging interpolation to construct a three-dimensional stress field, frequency domain processing adopts an adaptive spectral clustering method based on resonance peak tracking, and a nonlinear correlation matrix of time-space-frequency features is established through the graph attention network (GAT) to form a dynamically updated feature topology network.

[0013] Preferably, the hybrid anomaly detection module adopts a three-stage detection architecture: the first stage uses an improved robust isolation forest that introduces Mahalanobis distance to quickly screen the high-dimensional feature space; the second stage uses a stacked deep autoencoder to construct a feature reconstruction error surface, and identifies potential anomalies through dynamic threshold segmentation; the third stage uses a Bayesian inference network to fuse environmental parameters, historical cases and real-time monitoring data to generate anomaly judgment results with confidence ratings.

[0014] Preferably, the visual interactive interface integrates virtual reality technology and supports: GPU-accelerated dynamic rendering of the three-dimensional isosurface of the blasting vibration field (update frequency ≥ 30fps), mixed visualization of parallel coordinates and polar coordinates of multi-measurement point signals, and tracing of the spatiotemporal propagation path of abnormal events (supporting multi-time scale scaling), and gesture-driven feature dimension screening and profile analysis through the Leap Motion controller to form an immersive diagnostic environment.

[0015] Preferably, the self-learning optimization module includes a dual closed-loop feedback mechanism, a short-term incremental learning loop fine-tuning the model parameters through online stream data processing, a long-term model reconstruction loop performing full data retraining every 72 hours, and an online knowledge distillation technology based on temperature regulation to achieve a smooth transition between the old and new models. A feature weight adaptive adjustment unit based on Shapley value is also provided to dynamically optimize the feature selection strategy.

[0016] Preferably, the early warning and disposal module is configured with a hierarchical early warning strategy based on risk entropy value, constructs a comprehensive risk assessment model of geological conditions, charging parameters and environmental sensitivity through fuzzy Petri nets, designs a multimodal alarm output interface of sound, light, vibration and AR prompts, and integrates a case-based reasoning (CBR) engine to match the historical disposal solution library, and generates optimized disposal suggestions based on the current working conditions.

[0017] Preferably, the extensible interface module adopts a microservice architecture design, including: data access middleware supporting MQTT / OPC-UA dual protocols, an algorithm container interface integrating the TensorRT acceleration engine, a device communication interface compliant with the IEC 62541 standard, and an audit log evidence unit based on Hyperledger Fabric to achieve trusted traceability of operation traces.

[0018] The technical effects and advantages of the present invention are as follows:

[0019] 1. In response to the insufficient multi-source heterogeneous data fusion capabilities of existing technologies, the present invention innovatively adopts the WPT-EMD joint denoising algorithm and the dynamic normalization unit to work together. Through the fine frequency band division of wavelet packet transform (WPT) and the adaptive characteristics of empirical mode decomposition (EMD), it effectively overcomes the modal aliasing and noise residual problems of traditional methods. Combined with the dynamic normalization strategy based on information entropy, the feature space alignment of heterogeneous data such as vibration waveforms and stress fields is achieved. This design improves the signal-to-noise ratio of the original data and improves the consistency of the feature dimensions, laying a high-quality data foundation for subsequent analysis.

[0020] 2. To address the generalization bottleneck of existing anomaly detection technologies, a three-level hybrid detection model architecture is constructed. The improved robust isolation forest introduces the Mahalanobis distance metric to enhance the initial screening of anomalies in high-dimensional spaces; the stacked deep autoencoder captures dynamic patterns through a temporal attention mechanism; and the Bayesian inference network integrates 12 types of environmental parameters with a historical case library to form a multi-dimensional evidence chain.

[0021] 3. Addressing the weaknesses of existing technologies in human-machine collaborative decision-making, this invention develops an immersive intelligent interactive system. This system utilizes GPU-accelerated 3D isosurface rendering (30fps+) and gesture interaction technology to achieve holographic visualization of the blasting process. Furthermore, a dynamic early warning threshold system is constructed, combining a fuzzy Petri net risk assessment model with a case-based reasoning engine, enabling operators to intuitively understand changes in underground stress gradients.

[0022] 4. To address the shortcomings of existing systems in scalability and self-optimization, this invention builds a resilient and evolvable system architecture, employing a microservice containerized design and a dual-loop learning mechanism. Standardized interfaces support plug-and-play integration with new devices, reducing algorithm update time from hours to minutes. Online knowledge distillation technology enables hot model updates, ensuring 24 / 7 continuous operation. A Shapley value feature optimization module dynamically tracks operating condition changes, continuously adapting to the digital upgrade needs of blasting engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] As attached Figure 1As shown, a multi-dimensional analysis and anomaly detection processing system for blasting data is disclosed, including: a distributed data acquisition module, a data preprocessing module, a multi-dimensional analysis engine, a hybrid anomaly detection module, a visual interactive interface, an early warning and disposal module, and an extensible interface module; the distributed data acquisition module is connected to a vibration sensor array, a geological radar, and an unmanned aerial survey device through a multi-protocol adapter, the data preprocessing module establishes a unified spatiotemporal benchmark for the collected raw data, the multi-dimensional analysis engine maps the processed data into a feature space to construct a dynamic knowledge graph, the hybrid anomaly detection module implements cross-dimensional reasoning based on the node relationship of the knowledge graph, the visual interactive interface and the self-learning optimization module form a human-machine collaborative decision-making loop, the early warning and disposal module connects to the emergency control system to form a disposal closed loop, and the extensible interface module provides standardized expansion channels for each component, wherein, multi-source equipment such as vibration sensors and geological radars are integrated through a multi-protocol adapter to construct a heterogeneous data pool with a unified spatiotemporal benchmark. Its core advantage lies in achieving full-factor acquisition and high-fidelity conversion of blasting monitoring data, breaking through the data silos of traditional systems. Its distributed architecture uses edge computing nodes to process raw signals locally, reducing data transmission latency by over 80%. Its adaptive noise suppression unit eliminates environmental interference through a combined WPT-EMD algorithm, preserving effective signal characteristics. This design keeps the synchronization error of multi-point data at kilometer-level blasting sites within 5ms, providing a high-precision data foundation for subsequent analysis.

[0026] As a preferred embodiment, the data preprocessing module adopts a wavelet packet transform-empirical mode decomposition (WPT-EMD) joint denoising algorithm, processes non-stationary noise by constructing an adaptive threshold function, and is provided with a dynamic normalization unit based on feature importance. The unit calculates the information entropy weight of each measuring point through a sliding time window, implements nonlinear normalization on the vibration waveform data, and adopts distribution alignment based on kernel density estimation for discrete stress monitoring data to achieve feature space consistency conversion of heterogeneous data. Furthermore, the frequency band subdivision capability of the wavelet packet transform and the adaptive decomposition characteristics of the EMD are integrated to achieve non-stationary signal purification by constructing a dual threshold function (hard threshold to process high-frequency noise, soft threshold to smooth low-frequency fluctuations). The dynamic normalization unit uses a sliding window to calculate the information entropy weight, implements segmented Z-score normalization on the vibration data, and performs kernel density distribution alignment on discrete stress values. Actual tests show that this module improves the signal-to-noise ratio of the vibration waveform by 42% and the feature consistency of the stress data by 68%, effectively solving the feature distortion problem caused by traditional methods.

[0027] As a preferred embodiment, the multi-dimensional analysis engine is implemented through multimodal feature fusion technology: time domain feature extraction adopts the improved HHT transform that integrates the Marginal Spectrum entropy value, spatial domain analysis adopts the Delaunay triangulation algorithm combined with Kriging interpolation to construct a three-dimensional stress field, frequency domain processing adopts the adaptive spectral clustering method based on resonance peak tracking, and the nonlinear correlation matrix of time-space-frequency features is established through the graph attention network (GAT) to form a dynamically updated feature topology network. Furthermore, the marginal spectral entropy features are extracted by the improved HHT transform, the three-dimensional stress field model is constructed by Delaunay triangulation in the spatial domain, the resonance mode is identified by adaptive spectral clustering in the frequency domain, and the graph attention network (GAT) is used to establish the time-space-frequency feature correlation matrix to form a dynamically updated blasting knowledge graph. This technology increases the amount of effective information extracted from the original data by the feature dimension by 3 times, successfully captures 75% of the implicit correlation features missed by traditional methods, and provides three-dimensional feature support for anomaly detection;

[0028] As a preferred embodiment, the hybrid anomaly detection module adopts a three-stage detection architecture: the first stage uses an improved robust isolation forest that introduces Mahalanobis distance to quickly screen the high-dimensional feature space; the second stage uses a stacked deep autoencoder to construct a feature reconstruction error surface and identifies potential anomalies through dynamic threshold segmentation; the third stage uses a Bayesian inference network to fuse environmental parameters, historical cases and real-time monitoring data to generate anomaly judgment results with confidence ratings. Furthermore, the first-stage improved isolation forest introduces the Mahalanobis distance metric to solve the problem of high-dimensional space distance calculation distortion, and the initial screening efficiency reaches 5,000 data points per second; the second-stage stacked autoencoder improves the reconstruction accuracy to 97.3% through the temporal attention mechanism, and the dynamic threshold segmentation algorithm accurately identifies potential anomalies; the third-stage Bayesian network fuses 12 types of environmental parameters to generate a judgment result with a confidence level ≥ 0.95. The three-stage joint inspection mechanism makes the comprehensive detection accuracy reach 98.6%, which is 60% lower than the false alarm rate of the single model method, especially for slowly changing anomalies, the detection time is 2-3 sampling cycles earlier;

[0029] As a preferred embodiment, the visual interactive interface integrates virtual reality technology and supports: GPU-accelerated dynamic rendering of the blasting vibration field's three-dimensional isosurface (update frequency ≥ 30fps), hybrid visualization of parallel coordinates and polar coordinates of multi-measurement point signals, and backtracking of the spatiotemporal propagation path of abnormal events (supporting multi-time scale scaling). It also implements gesture-driven feature dimension screening and profile analysis through the Leap Motion controller to form an immersive diagnostic environment. Furthermore, GPU parallel computing is used to achieve real-time rendering of the blasting field's three-dimensional isosurface (frame rate ≥ 30fps), support gesture-interactive spatiotemporal dimension drill-down analysis, and hybrid visualization technology of parallel coordinates and polar coordinates, allowing operators to simultaneously observe the multi-dimensional feature evolution of more than 8 measurement points. The abnormal propagation path backtracking function combined with the particle tracking algorithm can simulate the abnormal diffusion rate and impact range. Actual measurements show that this interface increases data analysis efficiency by 4 times, decision response speed by 80%, and training cycle by 60%.

[0030] As a preferred embodiment, the self-learning optimization module includes a dual closed-loop feedback mechanism. The short-term incremental learning loop fine-tunes the model parameters through online stream data processing, and the long-term model reconstruction loop performs full data retraining every 72 hours. It uses online knowledge distillation technology based on temperature regulation to achieve a smooth transition between the old and new models, and is provided with a feature weight adaptive adjustment unit based on Shapley value to dynamically optimize the feature selection strategy. Furthermore, the short-term incremental learning loop updates the model parameters in real time through online stream processing (delay <1s), and the long-term model reconstruction loop performs full training every 72 hours to ensure generalization ability. The knowledge distillation technology adopts a temperature coefficient adjustment strategy to control the detection fluctuation during the transition between the old and new models within ±0.3%. The Shapley value feature optimization module dynamically adjusts the 32-dimensional feature weights, and the model adaptive adjustment time when the working conditions change is shortened from hours to minutes, ensuring that the system is continuously in the optimal state.

[0031] As a preferred embodiment, the early warning and disposal module is equipped with a hierarchical early warning strategy based on risk entropy. A comprehensive risk assessment model for geological conditions, charge parameters, and environmental sensitivity is constructed using fuzzy Petri nets. A multimodal alarm output interface is designed, featuring sound, light, vibration, and AR prompts. A case-based reasoning (CBR) engine is integrated to match a historical disposal solution library, generating optimized disposal recommendations based on current operating conditions. Furthermore, a five-level risk assessment model (continuously quantized risk entropy values ​​of 0-1) is constructed based on fuzzy Petri nets, integrating 18 influencing factors such as geological parameters and charge quantity. The case-based reasoning engine achieves second-level historical solution matching (with a library of over 5,000 cases), generating disposal recommendations with an accuracy rate of 92%. The multimodal alarm interface supports real-time overlay of risk hotspots using AR glasses, shortening on-site emergency response time to within 30 seconds and increasing the accuracy of major accident warnings to 99.2%.

[0032] As a preferred embodiment, the extensible interface module adopts a microservice architecture design, including: data access middleware supporting MQTT / OPC-UA dual protocols, an algorithm container interface integrating the TensorRT acceleration engine, a device communication interface compliant with the IEC62541 standard, and an audit log evidence unit based on Hyperledger Fabric to achieve trusted traceability of operation traces. Furthermore, the microservice containerized deployment supports dynamic loading of algorithm modules (startup time <2s), the MQTT / OPC-UA dual protocol adapter shortens the new device access cycle from weeks to hours, and the blockchain audit unit adopts the Hyperledger Fabric framework to achieve tamper-proof evidence of operation logs (processing 200+ transactions per second). This architecture enables the system to maintain 95% of real-time processing capabilities when the monitoring points are expanded by 50%, and the operation and maintenance costs are reduced by 45%, supporting the upgrade needs of blasting monitoring technology in the next ten years.

[0033] Example 1: Application of large-scale open-pit mine blasting monitoring scenario

[0034] 1. System architecture deployment

[0035] This system was deployed in an iron ore blasting operation area. The hardware configuration includes: a distributed data acquisition terminal with a 32-node vibration sensor array (1000Hz sampling rate), 8 geological radars (operating frequency 100MHz), and 2 drones (equipped with multispectral cameras); the edge computing nodes are NVIDIA Jetson AGX Xavier × 6, deployed with a data preprocessing module; the central server is a dual-core Xeon Gold 6230 processor + 4 × RTX A6000 GPU; the interactive terminals are HTC VIVE Pro 2 VR helmets × 3 and touch-screen command screens × 2; the network architecture uses a 5G private network with fiber optic redundancy backup to ensure data transmission delay is less than 20ms.

[0036] 2. Data preprocessing implementation

[0037] The scene data includes vibration waveform data: 32 channels × 120 seconds / blast (a total of 3.84 × 10^6 sampling points); geological radar data: 8 channels × 50m profile scan (resolution 0.1m); drone imagery: 2GB / blast multispectral point cloud data. First, WPT-EMD joint denoising: the vibration signal is first subjected to 5-layer wavelet packet decomposition, the db8 wavelet basis is selected, and the high-frequency coefficient (>250Hz) is processed with a hard threshold (λ = 3σ) to process the blasting impact noise; the low-frequency component is subjected to EMD decomposition, and the environmental vibration interference in the first 3-order IMF is discarded; the measured signal-to-noise ratio is improved from 15.6dB to 57.3dB, and dynamic normalization is performed. The information entropy weight is calculated using a sliding window (2 seconds), segmented Z-score normalization is implemented, and the Epanechnikov kernel function is used for distribution alignment, and the KL divergence is reduced by 72%

[0038] 3. Multi-dimensional feature analysis

[0039] Spatiotemporal feature extraction:

[0040] Time domain analysis: Improved HHT transform to extract marginal spectral entropy (window length 512 points, overlap rate 75%)

[0041] →Generate 32-dimensional time domain feature vector

[0042] Airspace modeling: Delaunay triangulation to construct three-dimensional stress field (interpolation accuracy ±0.15MPa)

[0043] →Identify three stress concentration areas (>8MPa)

[0044] Frequency domain clustering: Adaptive spectral clustering detected 4 main resonance peaks (85Hz / 210Hz / 430Hz / 680Hz)

[0045] Feature fusion:

[0046] Graph Attention Network (GAT) constructs a 128×128 correlation matrix

[0047] It is found that the nonlinear relationship between the temporal entropy value and the spatial stress gradient (correlation coefficient R 2 =0.87)

[0048] 4. Hybrid Anomaly Detection Process

[0049] An improved isolation forest (100 trees, 256 subsamplings) processed 320,000 data points in 0.8 seconds. The Mahalanobis distance threshold was set to χ2(0.99, 32) = 58.6, screening out 23 suspicious points. The stacked autoencoder (encoding layers 256-128-64, symmetric decoding) had a reconstruction error MAPE of 2.7%. Dynamic threshold segmentation (sliding window RMS comparison) identified five potential anomalies. A Bayesian network integrated geological parameters (rock mass strength RMR = 52) and historical cases (matching with 89% similarity) to output a "pre-crack surface instability" warning with a confidence level of 0.97.

[0050] 5. Visualization and Decision-making

[0051] 3D display, GPU-accelerated rendering of a blasting field 3D model (number of triangles > 5 million), with a stable frame rate of 45fps. Five fingers can be rotated to view the stress field gradient distribution, and a fist can be clenched and dragged on the time axis to retrace the anomaly propagation path (speed 0.5m / ms). The case-based reasoning engine matches three optimal solutions from a library of over 5,000 cases: adjusting the hole spacing to 4.2m, reducing the charge by 15%, and increasing the pre-crack hole density. AR glasses then mark high-risk areas in real time.

[0052] 6. Self-learning optimization

[0053] Double closed-loop update: Short-term loop: incremental update of model parameters every 5 minutes → dynamic adjustment of detection threshold by ±3.2% Long-term loop: full training every 72 hours (batch size 4096) → feature weight Shapley value update (maximum change weight from 0.15 to 0.22)

[0054] Knowledge distillation: The teacher model (ResNet-34) guides the student model (MobileNetV3). When the temperature coefficient τ = 3, the model switching detection fluctuation is less than 0.25%.

[0055] 7. Performance Verification Data

[0056] Detection accuracy: Missing alarm rate: 0.7%; False alarm rate: 1.2% (traditional method 19.6%)

[0057] Response speed from data collection to warning output: 3.8 seconds

[0058] System scalability: When 10 new monitoring points are added, processing delay only increases by 9%.

[0059] Effects of the embodiment

[0060] In the field test of this embodiment:

[0061] Successfully issued warnings for three major charging anomalies, optimized blasting parameters, and reduced the average bulk rate from 12% to 6.5%. The system has been running continuously for >2,000 hours without any troubles, and the operator training period has been shortened from 2 weeks to 3 days.

[0062] This implementation fully verifies the technical innovation and engineering practical value of the present invention in data fusion, intelligent detection, human-computer interaction and other aspects.

[0063] Finally, a few points should be explained: First, in the description of this application, it should be noted that, unless otherwise specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense, and may refer to mechanical or electrical connections, internal communication between two components, or direct connection. "Up," "down," "left," and "right" are only used to indicate relative positional relationships. When the absolute positions of the objects being described change, the relative positional relationships may also change.

[0064] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0065] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A blasting data multi-dimensional analysis and anomaly detection processing system, characterized in that: include: Distributed data acquisition module, data preprocessing module, multi-dimensional analysis engine, hybrid anomaly detection module, visual interactive interface, early warning and disposal module, and extensible interface module; The distributed data acquisition module connects to the vibration sensor array, geological radar and drone aerial survey equipment through a multi-protocol adapter. The data preprocessing module establishes a unified spatiotemporal benchmark for the collected raw data. The multi-dimensional analysis engine maps the processed data into the feature space to construct a dynamic knowledge graph. The hybrid anomaly detection module implements cross-dimensional reasoning based on the node relationship of the knowledge graph. The visual interactive interface and the self-learning optimization module form a human-computer collaborative decision-making loop. The early warning and disposal module connects to the emergency control system to form a closed disposal loop. The extensible interface module provides standardized expansion channels for each component.

2. The blasting data multi-dimensional analysis and anomaly detection processing system according to claim 1, characterized in that: The data preprocessing module adopts a wavelet packet transform-empirical mode decomposition (WPT-EMD) joint denoising algorithm to process non-stationary noise by constructing an adaptive threshold function. It is also equipped with a dynamic normalization unit based on feature importance. This unit calculates the information entropy weight of each measuring point through a sliding time window, implements nonlinear normalization on the vibration waveform data, and adopts distribution alignment based on kernel density estimation for discrete stress monitoring data to achieve feature space consistency conversion of heterogeneous data.

3. The blasting data multi-dimensional analysis and anomaly detection processing system according to claim 1, characterized in that: The multi-dimensional analysis engine is implemented through multimodal feature fusion technology: time domain feature extraction adopts the improved HHT transform that integrates the MarginalSpectrum entropy value; spatial domain analysis adopts the Delaunay triangulation algorithm combined with Kriging interpolation to construct a three-dimensional stress field; frequency domain processing adopts an adaptive spectral clustering method based on resonance peak tracking, and a nonlinear correlation matrix of time-space-frequency features is established through the graph attention network (GAT) to form a dynamically updated feature topology network.

4. The blasting data multi-dimensional analysis and anomaly detection processing system according to claim 1, characterized in that: The hybrid anomaly detection module adopts a three-stage detection architecture: the first stage uses an improved robust isolation forest that introduces Mahalanobis distance to quickly screen the high-dimensional feature space; the second stage uses a stacked deep autoencoder to construct a feature reconstruction error surface and identifies potential anomalies through dynamic threshold segmentation; the third stage uses a Bayesian inference network to fuse environmental parameters, historical cases and real-time monitoring data to generate anomaly judgment results with confidence ratings.

5. The blasting data multi-dimensional analysis and anomaly detection processing system according to claim 1, characterized in that: The visual interactive interface integrates virtual reality technology and supports: GPU-accelerated dynamic rendering of the blasting vibration field's three-dimensional isosurface (update frequency ≥ 30fps), hybrid visualization of parallel and polar coordinates for multi-measurement-point signals, and tracing back the spatiotemporal propagation paths of abnormal events (supporting multi-time-scale scaling). It also implements gesture-driven feature dimension screening and profile analysis through the Leap Motion controller, creating an immersive diagnostic environment.

6. The blasting data multi-dimensional analysis and anomaly detection processing system according to claim 1, characterized in that: The self-learning optimization module includes a dual closed-loop feedback mechanism. The short-term incremental learning loop fine-tunes model parameters through online streaming data processing. The long-term model reconstruction loop performs full data retraining every 72 hours. It uses temperature-regulated online knowledge distillation technology to achieve a smooth transition between the old and new models. It also has a feature weight adaptive adjustment unit based on Shapley values ​​to dynamically optimize the feature selection strategy.

7. The blasting data multi-dimensional analysis and anomaly detection processing system according to claim 1, characterized in that: The early warning and disposal module is equipped with a hierarchical early warning strategy based on risk entropy value. It constructs a comprehensive risk assessment model of geological conditions, charging parameters and environmental sensitivity through fuzzy Petri nets, designs a multimodal alarm output interface with sound, light, vibration and AR prompts, and integrates a case-based reasoning (CBR) engine to match the historical disposal solution library, generating optimized disposal suggestions based on the current working conditions.

8. The blasting data multi-dimensional analysis and anomaly detection processing system according to claim 1, characterized in that: The extensible interface module adopts a microservice architecture design and includes: data access middleware that supports MQTT / OPC-UA dual protocols, an algorithm container interface that integrates the TensorRT acceleration engine, a device communication interface that complies with the IEC 62541 standard, and an audit log notarization unit based on Hyperledger Fabric to achieve trusted traceability of operation traces.

Images