Strip mine slope landslide real-time prediction system
By combining two-channel data acquisition, edge computing, and Fourier transform with Kalman filtering, the problems of slow data acquisition, insufficient feature extraction, and simplistic model construction in open-pit mine slope monitoring were solved, achieving efficient and accurate real-time landslide prediction and early warning.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-30
- Publication Date
- 2026-03-27
AI Technical Summary
Existing open-pit mine slope monitoring technologies suffer from low data acquisition efficiency and susceptibility to interference, insufficient feature extraction, poor multi-sensor data processing and fusion, simplistic model construction, and unreasonable early warning standards, resulting in insufficient accuracy in real-time landslide prediction and early warning.
The system employs dual-channel data acquisition, edge computing gateways to compress data, and Ethernet transmission. It combines Fourier transform to extract dynamic change features, uses Kalman filtering to fuse multi-sensor data, establishes a weighted voting method for fusion, and establishes a physical and data-driven model to divide the data into multiple levels of early warning standards, thereby achieving efficient and accurate prediction.
It improved data collection efficiency and anti-interference ability, enhanced the comprehensiveness of feature extraction and the accuracy of data processing, improved the prediction accuracy and early warning timeliness of the model, reduced the false alarm and false negative rates, and reduced economic losses.
Smart Images
Figure CN121747273A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of slope and landslide technology, specifically relating to a real-time prediction system for open-pit mine slope landslides. Background Technology
[0002] Traditional open-pit mine slope monitoring and prediction technologies suffer from significant shortcomings in multiple aspects, making it difficult to meet the demands for real-time, accurate, and economical safety monitoring. In terms of data acquisition, single-channel sensors have low sampling rates, failing to capture rapid deformations that are precursors to landslides; their anti-interference capabilities are weak, with electromagnetic interference in mining areas causing data noise to account for over 30%, resulting in a signal-to-noise ratio (SNR) of <10dB; transmission bandwidth is severely wasted, as raw data is uncompressed, requiring a total bandwidth of >500KB / s from multiple sensors, far exceeding wireless transmission capabilities, leading to a packet loss rate of 20%-40%; wired transmission reliability is poor, with complex terrain in mining areas causing an average annual line damage rate of 15%-25%, and maintenance costs accounting for over 30%. In the feature extraction stage, only static features are extracted, ignoring dynamic change rates (such as sudden changes in displacement acceleration) and frequency domain features, resulting in delayed identification of landslide precursors; large time alignment errors (0.1-1s) between multiple sensors reduce feature fusion accuracy by more than 20%. In terms of data processing, the multi-sensor fusion method is crude, failing to consider correlations, resulting in high variance of fused data (40%-60%); noise suppression is insufficient, with residual standard deviations reaching 0.5-1 mm; time-series data is not compressed for storage, with annual data volume of 10-20 GB per sensor and query latency >5 seconds. In the model building and early warning stages, physical model parameter calibration errors are large (20%-30%), data-driven models have poor generalization ability (accuracy drops by 15%-25% with geological changes), and overall prediction errors reach 20%-40%; early warning standards use fixed thresholds, failing to consider geological differences, resulting in false alarm rates of 15%-25%, false negative rates of 10%-20%, and annual economic losses exceeding one million yuan.
[0003] In existing open-pit mine slope monitoring technologies, data acquisition is inefficient due to its single-channel nature and is susceptible to interference during transmission; data processing and fusion are poor and storage is inefficient; model building lacks fusion strategies; and early warning standards are unscientific, leading to frequent false alarms and missed alarms. Summary of the Invention
[0004] To address the aforementioned issues, this invention provides a real-time landslide prediction system for open-pit mine slopes. This system solves the problems of existing technologies, such as slow and easily interfered data acquisition, insufficient feature extraction and ineffective coordination, poor multi-sensor data processing and fusion, inaccurate and simplistic model construction, lack of in-depth analysis in predictions, and unreasonable early warning standards, thus failing to achieve efficient and accurate real-time landslide prediction and early warning. To achieve the above objectives, this invention adopts the following technical solution: The aforementioned real-time prediction system for landslides on open-pit mine slopes includes the following modules: a data acquisition module for acquiring monitoring data, which is collected simultaneously through two channels, compressed using an edge computing gateway, and transmitted via Ethernet to obtain acquired data; a feature extraction module for acquiring displacement, rainfall, and temporal features from the acquired data, capturing dynamic trends, extracting the energy proportion of the main frequency band using Fourier transform, and aligning the temporal, frequency, and spatial features using timestamps to obtain feature data; a data processing module for acquiring the feature data from the feature extraction module, fusing multi-sensor data using Kalman filtering, and storing the processed data in a time-series database to obtain processed data; a model building module for acquiring the processed data, establishing a physical model and a data-driven model, and fusing the physical model and the data-driven model using a weighted voting method to obtain a digital model; a prediction module for acquiring the digital model, evaluating the model construction using a confusion matrix and ROC curve, generating a weekly prediction report, summarizing the characteristics of high-risk periods, and obtaining prediction data; and an early warning module for acquiring the prediction data, defining early warning standards, and pushing early warning alarms when the risk score exceeds the requirements of the early warning standards.
[0005] Furthermore, the data acquisition module includes the following units: a data acquisition unit for acquiring monitoring data on slope displacement, vibration, groundwater level, soil moisture content, rainfall, and slope stress; a data synchronization acquisition unit for simultaneous acquisition by dual-channel data acquisition instruments, employing a built-in watchdog program to ensure automatic recovery after abnormal power outages; a data pre-transmission unit for preliminary verification of monitoring data, removing obvious erroneous values, adding timestamps and sensor ID metadata, and using an edge computing gateway for data compression and encryption; and a transmission unit for transmitting the acquired data via industrial Ethernet and the MQTT transmission protocol.
[0006] Furthermore, the feature extraction module includes the following units: a dynamic change unit, used to acquire displacement, rainfall, and temporal features from the collected data, and obtain the dynamic change trend by calculating displacement rate, acceleration, cumulative displacement, 3-day cumulative rainfall, maximum hourly rainfall intensity, and root mean square value and peak factor of the vibration signal; a main frequency unit, used to perform fast Fourier transform on the vibration signal to obtain the main frequency and the energy proportion of the 10 to 50 Hz frequency band, and identify the characteristic frequency bands of blasting vibration and rock mass fracture; and a normalization processing unit, used to align the temporal, frequency domain, and spatial features by timestamp, construct feature vectors, and use normalization processing to unify the input format to obtain feature data.
[0007] Furthermore, the data processing module includes the following units: a dimensionality reduction unit, used to acquire feature data from the feature extraction module, fuse the data using Kalman filtering technology, and reduce the dimensionality of high-dimensional features using PCA to obtain processed data; and a data processing unit, used to store the processed data into a time-series database, cache data from the most recent 24 hours, and obtain processed data.
[0008] Furthermore, the model building module includes the following units: a physical model unit, used to acquire and process data, and construct a slope stability analysis model using the limit equilibrium method to obtain a physical model; a data-driven model unit, used to capture displacement time series using an LSTM network, and construct an XGBoost classification model to obtain a data-driven model; and a digital model unit, used to fuse the physical model and the data-driven model using a weighted voting method, and obtain a digital model through Bayesian optimization.
[0009] Furthermore, the prediction module includes the following units: an evaluation unit, used to acquire a digital model, calculate the accuracy, false alarm rate, false negative rate and ROC curve of the confusion matrix, and evaluate the performance of the digital model; and a prediction data unit, used to generate a prediction report weekly, summarize the characteristics of high-risk periods, optimize sensor deployment strategies, and obtain prediction data.
[0010] Furthermore, the early warning module includes the following units: an early warning standard unit, used to acquire predictive data and divide the early warning standard into three levels: yellow warning, orange warning and red warning, used to push an early warning alarm when the risk score exceeds the requirements of the early warning standard.
[0011] Furthermore, the feature extraction module is used to acquire displacement, rainfall, and temporal features from the collected data. By capturing dynamic trends, Fourier transform is used to extract the energy proportion of the main frequency band. The temporal, frequency, and spatial features are aligned with timestamps to obtain feature data, including: obtaining temporal features by calculating displacement rate, acceleration, and 3-day cumulative rainfall; obtaining frequency domain features by performing FFT transform on the vibration signal to extract the main frequency and the energy proportion of the 10 to 50 Hz frequency band; and obtaining spatial features by combining the slope-height product with the DEM model to identify high-risk areas.
[0012] Furthermore, the early warning standard unit includes the following sub-units: a yellow warning sub-unit, used to acquire displacement rate and landslide probability data; when the displacement rate is 3 to 5 mm / d or the landslide probability is 0.3 to 0.5, a yellow warning is met, and the safety officer is notified via SMS; an orange warning sub-unit, used to acquire displacement rate and landslide probability data; when the displacement rate is 5 to 8 mm / d or the landslide probability is 0.5 to 0.7, an orange warning is met, and the audible and visual alarms are activated and patrols are strengthened; a red warning sub-unit, used to acquire displacement rate and landslide probability data; when the displacement rate is greater than or equal to 8 mm / d or the landslide probability is greater than or equal to 0.7, a red warning is met, and the danger zone is highlighted through linkage with the GIS system, and emergency calls are automatically dialed.
[0013] In the technical solution provided by this invention, a data acquisition module is used to acquire monitoring data. This data is acquired simultaneously through two channels, compressed using an edge computing gateway, and transmitted via Ethernet to obtain the acquired data. A feature extraction module is used to acquire displacement, rainfall, and temporal features from the acquired data. By capturing dynamic trends, Fourier transform is used to extract the energy proportion of the main frequency band. The temporal, frequency, and spatial features are aligned using timestamps to obtain feature data. A data processing module is used to acquire the feature data from the feature extraction module. Kalman filtering is used to fuse multi-sensor data, and the processed data is stored in a time-series database to obtain processed data. A model building module is used to acquire the processed data, establish a physical model and a data-driven model, and fuse the physical model and the data-driven model using a weighted voting method to obtain a digital model. A prediction module is used to acquire the digital model, evaluate the model construction using a confusion matrix and ROC curve, generate a weekly prediction report, summarize the characteristics of high-risk periods, and obtain prediction data. An early warning module is used to acquire the prediction data, define early warning standards, and push an early warning alarm when the risk score exceeds the early warning standard requirements. This invention solves the problems of existing technologies, such as slow and easily interfered data acquisition, insufficient and ineffective feature extraction and fusion, poor multi-sensor data processing and fusion, single and inaccurate model construction, lack of in-depth analysis in prediction, and unreasonable early warning standards, which prevent efficient and accurate real-time prediction and early warning of landslides. Attached Figure Description
[0014] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.
[0015] Figure 1 This is a schematic diagram of the first embodiment of a real-time prediction system for landslides on open-pit mine slopes according to an embodiment of the present invention.
[0016] Figure 2This is a schematic diagram of a second embodiment of a real-time prediction system for landslides on open-pit mine slopes according to an embodiment of the present invention.
[0017] Figure 3 This is a schematic diagram of a third embodiment of a real-time prediction system for landslides on open-pit mine slopes according to an embodiment of the present invention.
[0018] Figure 4 This is a schematic diagram of the fourth embodiment of a real-time prediction system for landslides on open-pit mine slopes in this invention.
[0019] Figure 5 This is a schematic diagram of the fifth embodiment of a real-time prediction system for landslides on open-pit mine slopes in this invention. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.
[0021] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, modules, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, modules, operations, elements, components, and / or groups thereof.
[0022] A real-time prediction system for landslides on open-pit mine slopes, such as Figure 1 As shown, the system includes the following modules: a data acquisition module, used to acquire monitoring data simultaneously through two channels, compress the data using an edge computing gateway, and transmit it via Ethernet to obtain the acquired data; a feature extraction module, used to acquire displacement, rainfall, and temporal features from the acquired data, capturing dynamic trends, using Fourier transform to extract the energy proportion of the main frequency band, and aligning the temporal, frequency, and spatial features using timestamps to obtain feature data; a data processing module, used to acquire the feature data from the feature extraction module, using Kalman filtering to fuse multi-sensor data, and storing the processed data in a time-series database to obtain processed data; a model building module, used to acquire the processed data, establish a physical model and a data-driven model, and use a weighted voting method to fuse the physical model and the data-driven model to obtain a digital model; a prediction module, used to acquire the digital model, evaluate the model construction using a confusion matrix and ROC curve, generate a weekly prediction report, summarize the characteristics of high-risk periods, and obtain prediction data; and an early warning module, used to acquire the prediction data, define early warning standards, and push early warning alarms when the risk score exceeds the early warning standard requirements.
[0023] To acquire monitoring data, the system simultaneously collects data through two channels (displacement sensor channel and multi-parameter sensor channel), compresses the raw data using an edge computing gateway (compression rate ≥70%), and transmits it via Ethernet with low latency (<50ms) to obtain high-precision acquired data.
[0024] Specifically, Kalman filtering is an algorithm that uses the state equations of a linear system and observes the system's input and output data to optimally estimate the system's state. The advantages of Kalman filtering include its suitability for constantly changing systems, low memory footprint, high speed, and suitability for embedded systems. It can extract useful information from uncertainties and make informed predictions about the system's next steps, even in situations with significant uncertainty. Kalman filtering has been widely applied in communication, navigation, guidance, and control, and is indispensable in algorithms such as lane following, target tracking, and radar tracking in autonomous driving.
[0025] The core of Kalman filtering is the state equation and observation equation of a linear system. The specific state equation is as follows: ,in Let A be the system state vector at time k (e.g., position, velocity, acceleration, etc.), A be the state transition matrix (describing the linear evolution of the state over time; for example, it is the identity matrix for uniform motion, and includes the relationship between velocity and acceleration for uniformly accelerated motion), and B be the control input matrix. Process noise. The observation equation is: ; Let be the observed value at time k (such as displacement, velocity, etc. measured by the sensor); H is the observation matrix. Observational noise.
[0026] like Figure 2 As shown, in this embodiment, the data acquisition unit is used to acquire monitoring data on slope displacement, vibration, groundwater level, soil moisture content, rainfall, and slope stress; the data synchronization acquisition unit is used for simultaneous acquisition by dual-channel data acquisition instruments, employing a built-in watchdog program to ensure automatic recovery after abnormal power outages; the data pre-transmission unit is used to perform preliminary verification of the monitoring data, remove obvious erroneous values, add timestamps and sensor ID metadata, and use an edge computing gateway for data compression and encryption; the transmission unit is used to transmit the acquired data via industrial Ethernet and the MQTT transmission protocol.
[0027] The data acquisition unit can comprehensively acquire various key monitoring data such as slope displacement and vibration, providing a rich information foundation for subsequent analysis. The synchronous data acquisition unit utilizes dual-channel data acquisition instruments operating simultaneously to improve acquisition efficiency. A built-in watchdog program automatically resumes acquisition in the event of a power outage, ensuring continuous data acquisition. The data pre-transmission unit performs preliminary verification of the acquired data, eliminating obvious errors to ensure data quality; it adds timestamps and sensor ID metadata for easy data traceability and management; and it uses an edge computing gateway to compress and encrypt data, reducing the amount of data transmitted and enhancing data security. The transmission unit employs industrial Ethernet and MQTT transmission protocols. Industrial Ethernet ensures stable transmission, while the MQTT protocol makes transmission more efficient and consumes less bandwidth.
[0028] The coordinated operation of each unit enables efficient, reliable, and secure collection and transmission of slope monitoring data, providing high-quality data support for subsequent data processing, model building, and landslide prediction. This helps improve the accuracy and timeliness of landslide prediction for open-pit mine slopes, and reduces safety risks and economic losses.
[0029] A watchdog timer is a widely used software or hardware mechanism in computers and embedded systems. Its main function is to monitor the system's operating status and prevent abnormal situations such as system crashes or freezes. Watchdog timers are characterized by their independence and high reliability; they typically run independently of the main program and are unaffected by main program failures. In fields with extremely high system stability requirements, such as embedded systems, industrial control, and aerospace, watchdog timers are one of the key technologies for ensuring reliable system operation, effectively improving system fault tolerance and stability, and reducing losses caused by system failures.
[0030] like Figure 3 As shown in this embodiment, the dynamic change unit is used to acquire displacement, rainfall, and temporal characteristics from the collected data. By calculating displacement rate, acceleration, cumulative displacement, 3-day cumulative rainfall, maximum hourly rainfall intensity, and root mean square value and peak factor of the vibration signal, the dynamic change trend is captured. The main frequency unit is used to perform fast Fourier transform on the vibration signal to obtain the main frequency and the energy proportion of the 10 to 50 Hz frequency band, and to identify the characteristic frequency bands of blasting vibration and rock mass fracture. The normalization processing unit is used to align the temporal, frequency domain, and spatial features according to the timestamp, construct feature vectors, and use normalization processing to unify the input format to obtain feature data.
[0031] The dynamic change unit can keenly capture the dynamic trend of data changes by acquiring displacement, rainfall, and temporal characteristics, and calculating key indicators such as displacement rate. This helps to promptly detect abnormal displacement changes and the impact of rainfall on the system, providing data support for early warning of potential risks and making response measures more forward-looking. The main frequency unit performs a fast Fourier transform on the vibration signal, extracting the energy proportion of the main frequency band (e.g., the energy proportion of the low-frequency band of 0.5-5Hz reflects rock mass creep, and the energy proportion of the high-frequency band of 10-50Hz reflects the precursor of rupture) to obtain the main frequency and the energy proportion of specific frequency bands, which can accurately identify the characteristic frequency bands of blasting vibration and rock mass rupture. This provides a scientific basis for judging the stability of rock mass and the blasting effect, helps to understand the internal condition of rock mass, and ensures engineering safety. The normalization processing unit aligns different types of features by timestamp to construct feature vectors, unifies all feature data to the UTC time base with a time synchronization error of <0.1s, and performs normalization processing to unify the input format. This approach eliminates the differences in dimensions and orders of magnitude among features, allowing each feature to participate fairly in the analysis, improving the accuracy of data analysis and the stability of model training, and laying a solid foundation for subsequent efficient and accurate data processing and decision-making.
[0032] Traditional methods extract only static features (such as maximum displacement and cumulative rainfall), ignoring dynamic rates of change (such as displacement acceleration and rainfall intensity variation). For example, before slope instability, displacement acceleration may suddenly increase from 0.1 mm / min² to 1 mm / min², but traditional methods cannot quantify such abrupt changes. Furthermore, they only analyze time-domain signals, failing to extract frequency-domain features (such as the energy proportion of the dominant frequency band) through Fourier transform or wavelet transform. Studies show that before slope instability, the dominant frequency of vibration signals may shift from 1-5 Hz to 10-20 Hz, but traditional methods cannot identify such frequency domain shifts.
[0033] The Fast Fourier Transform (FFT) is a practical algorithm for efficiently calculating the Discrete Fourier Transform (DFT). The DFT transforms signals from the time domain to the frequency domain, revealing the frequency components, but traditional methods are computationally intensive. This algorithm cleverly utilizes the periodicity and symmetry of the DFT calculation, employing a divide-and-conquer strategy to progressively decompose a long sequence of DFTs into multiple shorter sequences, finally merging the results using a butterfly operation. This algorithm significantly reduces computational complexity and greatly improves speed, making frequency domain analysis, which previously required substantial time and resources, highly efficient and feasible. It has crucial applications in numerous fields: in communications, it aids in signal modulation and demodulation; in audio processing, it enables spectral analysis and audio noise reduction; and in image processing, it can be used for image compression and feature extraction, providing powerful tools for the development of modern information technology.
[0034] like Figure 4 As shown in this embodiment, the dimensionality reduction unit is used to obtain feature data from the feature extraction module, fuse the data using Kalman filtering technology, and reduce the dimensionality of high-dimensional features using PCA to obtain the processed data; the data processing unit is used to store the processed data into a time series database, cache the data of the most recent 24 hours, and obtain the processed data.
[0035] Kalman filtering is employed to fuse multi-sensor data, eliminating noise (noise suppression rate ≥80%). The processed data is then stored in a time-series database (InfluxDB), supporting millisecond-level queries. These operations effectively improve data quality, reduce data processing complexity, and enhance the efficiency of data processing and storage. This provides more accurate and efficient data support for open-pit mine slope landslide prediction models, improving the timeliness and accuracy of predictions, thereby better ensuring the safety of open-pit mine production.
[0036] Traditional weighted averaging or simple superposition fusion does not consider sensor correlations (such as the strong coupling between displacement and pore water pressure), resulting in a 40%-60% higher variance in the fused data than the theoretical value, and poor interference removal. Furthermore, no filtering algorithm was designed specifically for the noise characteristics of mining areas (such as the high-frequency impact of blasting vibrations and the low-frequency jitter of mechanical noise), leading to a residual standard deviation of 0.5-1 mm after processing, far exceeding the landslide early warning threshold (typically 2-5 mm).
[0037] Specifically, PCA is a powerful data dimensionality reduction and feature extraction method. It can reduce data dimensionality, reduce computation and storage space, and improve data processing efficiency; eliminate correlation between features, remove data redundancy, and make data simpler and easier to analyze; it can also be used for data visualization, reducing high-dimensional data to two-dimensional or three-dimensional display.
[0038] like Figure 5 As shown, in this embodiment, the physical model unit is used to acquire and process data, construct a slope stability analysis model using the limit equilibrium method, and obtain the physical model; the data-driven model unit is used to capture displacement time series using an LSTM network, construct an XGBoost classification model, and obtain the data-driven model; the digital model unit is used to fuse the physical model and the data-driven model using a weighted voting method, and obtain the digital model through Bayesian optimization.
[0039] The physical modeling unit utilizes the limit equilibrium method to construct a slope stability analysis model using processed data. As a classic theoretical method, the limit equilibrium method provides an intuitive analysis of slope stability based on mechanical principles, offering a solid theoretical foundation and physical basis for the entire study and clarifying the safety status of slopes under various conditions. The data-driven modeling unit employs an LSTM network to capture displacement time-series features and combines it with an XGBoost classification model to construct a data-driven model. LSTM networks excel at processing time-series data and can uncover potential patterns in displacement changes; the XGBoost classification model can effectively classify slope conditions. The combination of these two methods enhances the model's ability to handle complex data and its predictive accuracy. The digital modeling unit integrates the physical model and the data-driven model through a weighted voting method, fully leveraging the advantages of both, and further improves model performance through Bayesian optimization. This fusion approach gives the digital model both theoretical rigor and data adaptability, enabling a more accurate and comprehensive assessment of slope conditions and providing reliable support for engineering decisions.
[0040] In this embodiment, the evaluation unit is used to acquire the digital model, calculate the accuracy, false alarm rate, false negative rate and ROC curve of the confusion matrix, and evaluate the performance of the digital model; the prediction data unit is used to generate a prediction report every week, summarize the characteristics of high-risk periods, optimize the sensor deployment strategy, and obtain prediction data.
[0041] Model performance is evaluated using a confusion matrix (accuracy ≥ 90%) and ROC curves (AUC ≥ 0.95). Weekly forecast reports are generated, summarizing the characteristics of high-risk periods (e.g., "displacement acceleration > 0.5 mm / min² after 48 hours of continuous rainfall").
[0042] In this embodiment, the early warning standard unit is used to acquire prediction data and divide the early warning standard into three levels: yellow warning, orange warning and red warning. It is used to push an early warning alarm when the risk score exceeds the requirements of the early warning standard.
[0043] Warning level classification and thresholds:
[0044] In this embodiment, the temporal feature unit is used to calculate the displacement rate, acceleration, and 3-day cumulative rainfall to obtain temporal features; the frequency domain feature unit is used to perform FFT transformation on the vibration signal to extract the main frequency and the energy proportion of the 10 to 50 Hz frequency band to obtain frequency domain features; and the spatial feature unit is used to calculate the slope-height product of the DEM model to identify high-risk areas to obtain spatial features.
[0045] In this embodiment, the yellow warning subunit is used to acquire displacement rate and landslide probability data. When the displacement rate is 3 to 5 mm / d or the landslide probability is 0.3 to 0.5, a yellow warning is triggered, and the safety officer is notified via SMS. The orange warning subunit is used to acquire displacement rate and landslide probability data. When the displacement rate is 5 to 8 mm / d or the landslide probability is 0.5 to 0.7, an orange warning is triggered, and audible and visual alarms are activated and patrols are intensified. The red warning subunit is used to acquire displacement rate and landslide probability data. When the displacement rate is greater than or equal to 8 mm / d or the landslide probability is greater than or equal to 0.7, a red warning is triggered, and the danger zone is highlighted through linkage with the GIS system, and emergency calls are automatically dialed.
[0046] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.
Claims
1. A real-time prediction system for slope failure in an open pit mine, characterized in that, The open-pit mine slope landslide real-time prediction system comprises the following modules: A data acquisition module is configured to acquire monitoring data, acquire the monitoring data through two-channel simultaneous acquisition, compress the data through an edge computing gateway, transmit the data through Ethernet, and obtain acquisition data. A feature extraction module is configured to acquire displacement, rainfall, and time sequence features in the acquisition data, capture dynamic change trends, extract main frequency band energy ratios through Fourier transform, align time sequence, frequency domain, and spatial features through timestamps, and obtain feature data. A data processing module is configured to acquire the feature data in the feature extraction module, fuse multi-sensor data through Kalman filtering, store the processed data in a time sequence database, and obtain data processing data. A model construction module is configured to acquire the processing data, construct a physical model and a data-driven model, fuse the physical model and the data-driven model through a weighted voting method, and obtain a digital model. A prediction module is configured to acquire the digital model, evaluate model construction through a confusion matrix and an ROC curve, generate a prediction report every week, summarize high-risk period features, and obtain prediction data. An early warning module is configured to acquire the prediction data, divide early warning standards, and push an early warning alarm when a risk score exceeds a requirement of the early warning standard.
2. The open-pit mine slope landslide real-time prediction system according to claim 1, characterized in that, The data acquisition module comprises the following units: A data acquisition unit is configured to acquire displacement, vibration, underground water level, soil moisture content, rainfall, and slope stress monitoring data of a slope. A data synchronous acquisition unit is configured to simultaneously acquire data through a double-channel data acquisition instrument, and ensure automatic recovery after abnormal power failure through an internal watchdog program. A data pre-transmission unit is configured to preliminarily check monitoring data, eliminate obvious error values, add timestamp and sensor ID metadata, compress and encrypt data through an edge computing gateway. A transmission unit is configured to transmit acquisition data through MQTT transmission protocol through industrial Ethernet.
3. The open-pit mine slope landslide real-time prediction system according to claim 1, characterized in that, The feature extraction module comprises the following units: A dynamic change unit is configured to acquire displacement, rainfall, and time sequence features in the acquisition data, capture dynamic change trends through calculation of displacement rate, acceleration, displacement cumulative amount, 3-day cumulative amount of rainfall data, maximum hourly rainfall intensity, and root mean square value and peak value factor of a vibration signal. A main frequency unit is configured to perform fast Fourier transform on a vibration signal, obtain main frequency and 10-50 Hz frequency band energy ratios, and identify blasting vibration and rock mass rupture feature bands. A normalization processing unit is configured to align time sequence, frequency domain, and spatial features through timestamps, construct a feature vector, uniformly input formats through normalization processing, and obtain feature data.
4. The open-pit mine slope landslide real-time prediction system according to claim 1, characterized in that, The data processing module comprises the following units: A dimension reduction unit is configured to acquire feature data in the feature extraction module, fuse data through Kalman filtering technology, and reduce the dimension of high-dimensional features through PCA to obtain processed data. A processing data unit is configured to store the processed data in a time sequence database, cache recent 24-hour data, and obtain processing data.
5. The open-pit mine slope landslide real-time prediction system according to claim 1, characterized in that, The model construction module comprises the following units: The physical model unit is used to acquire and process data, and to construct a slope stability analysis model using the limit equilibrium method to obtain the physical model; The data-driven model unit is used to capture displacement time series through an LSTM network, construct an XGBoost classification model, and obtain a data-driven model. The digital model unit is used to fuse the physical model and the data-driven model through a weighted voting method, and obtain the digital model through Bayesian optimization.
6. The open-pit mine slope landslide real-time prediction system according to claim 1, characterized in that, The prediction module includes the following units: The evaluation unit is used to acquire the digital model, calculate the accuracy, false positive rate, false negative rate and ROC curve of the confusion matrix, and evaluate the performance of the digital model. The predictive data unit is used to generate weekly predictive reports, summarize the characteristics of high-risk periods, optimize sensor deployment strategies, and obtain predictive data.
7. The open-pit mine slope landslide real-time prediction system according to claim 1, characterized in that, The early warning module includes the following units: The early warning standard unit is used to acquire forecast data and divide the early warning standard into three levels: yellow, orange and red. It is used to push early warning alarms when the risk score exceeds the requirements of the early warning standard.
8. The open-pit mine slope landslide real-time prediction system according to claim 1, characterized in that, The feature extraction module includes the following units: The time-series feature unit is used to calculate displacement rate, acceleration, and 3-day cumulative rainfall to obtain time-series features; The frequency domain feature unit is used to perform FFT transformation on the vibration signal, extract the main frequency and the energy ratio of the 10 to 50 Hz frequency band, and obtain the frequency domain features. Spatial feature units are used to calculate the slope-height product in the DEM model, identify high-risk areas, and obtain spatial features.
9. The open-pit mine slope landslide real-time prediction system according to claim 7, characterized in that, The early warning standard unit includes the following sub-units: The yellow alert subunit is used to acquire displacement rate and landslide probability data. When the displacement rate is 3 to 5 mm / d or the landslide probability is 0.3 to 0.5, a yellow alert is triggered, and the safety officer is notified by sending an SMS. The orange alert subunit is used to acquire displacement rate and landslide probability data. When the displacement rate is 5 to 8 mm / d or the landslide probability is 0.5 to 0.7, an orange alert is triggered by activating audible and visual alarms and strengthening patrols. The red alert subunit is used to acquire displacement rate and landslide probability data. When the displacement rate is greater than or equal to 8 mm / d or the landslide probability is greater than or equal to 0.7, a red alert is triggered. The danger zone is highlighted through linkage with the GIS system, and emergency calls are automatically dialed.