A mine real-time risk early warning system based on internet of things and artificial intelligence

Abstract

The scheme is composed of coupled sensing data group by matching the position sensor and the vibration sensor at the same physical point inside the slope. The six-dimensional feature vector is extracted, including vibration amplitude, vibration frequency, displacement accumulation, displacement rate, depth distribution index and time evolution index. The slope state is divided into four modes: construction vibration, surface slip, structure damage and safety. The Gaussian membership function is used to calculate the membership degree of each mode, and the transition state mechanism is introduced to solve the boundary fuzzy problem. The smooth identification of mode conversion is realized through the fourth-order state transition probability matrix and the hidden Markov model. Four typical evolution path templates are preset, including construction cycle, rainfall-induced landslide, deep rupture evolution and construction-induced damage. After dynamic time warping matching, the next stage mode and the remaining warning time are predicted. Combined with spatial distribution discrimination, environmental factor decoupling and confidence evaluation, the final output of disaster type and warning level is obtained.

Description

Technical Field

[0001] This invention relates to the field of mine early warning, and in particular to a real-time mine risk early warning system based on the Internet of Things and artificial intelligence. Background Technology

[0002] Slope instability is one of the most significant geological hazards in open-pit mines, seriously threatening mine production safety and the lives and property of personnel. With the development of IoT and sensor technologies, mine slope monitoring has gradually evolved from traditional manual inspections to automated online monitoring, with various sensors such as displacement, vibration, stress, and seepage pressure sensors being widely used for slope condition sensing.

[0003] In practical engineering applications, the core challenge of slope monitoring lies in the ambiguity of signals. Routine production activities such as construction blasting, mechanical operations, and vehicle transportation can all cause slope vibrations and instantaneous deformations. These signals are highly similar in amplitude characteristics to actual landslide precursor signals, making it highly susceptible to false alarms when relying solely on single-parameter thresholds. Existing monitoring systems generally use displacement rate or vibration amplitude as alarm thresholds. Frequent false alarms triggered during construction not only reduce system reliability but also lull maintenance personnel into complacency regarding genuine early warning signals.

[0004] A deeper problem lies in the fact that different types of slope hazards exhibit drastically different evolutionary mechanisms and signal characteristics. Surface soil creep is characterized by continuously increasing displacement but weak vibrations, while deep rock fracturing is characterized by frequent micro-vibrations and accelerated displacement developing simultaneously. Construction disturbances, on the other hand, are characterized by strong vibrations but complete displacement recovery. Most existing monitoring systems output a single risk index or risk level, failing to automatically distinguish between hazard types. Emergency responders struggle to take targeted measures based on alarm information, delaying optimal response times.

[0005] From a monitoring architecture perspective, existing systems typically deploy displacement and vibration sensors independently, with the two types of data then fused in the cloud or platform layer. Under this architecture, when deep, hidden cracks occur, displacement and vibration signals are difficult to capture simultaneously at the same physical location. The time and spatial differences between the signals reduce the sensitivity and accuracy of early identification. Furthermore, sensor arrays generally use fixed spacing, failing to consider the stress distribution gradient differences caused by variations in slope inclination. This results in insufficient monitoring density in high-stress areas and wasted resources in low-stress areas, necessitating further improvement in monitoring efficiency.

[0006] At the level of intelligent algorithms, deep learning models such as LSTM and TCN have been introduced into slope displacement prediction in recent years, improving prediction accuracy compared to traditional methods. However, these purely data-driven models lack mechanical constraints, easily producing predictions that are numerically accurate but physically unreasonable. Furthermore, their interpretability is poor, making it difficult for engineers to understand and trust the algorithm's output. In addition, most existing AI models focus on numerical displacement prediction, failing to delve into the coupling relationship between displacement and vibration, as well as the stage-specific characteristics of pattern evolution, resulting in insufficient depth and guidance in early warning information.

[0007] Therefore, how to automatically distinguish between construction vibrations and actual landslides, how to automatically identify slope disaster types, how to improve the early identification sensitivity of deep hidden ruptures, and how to make the early warning results physically interpretable are the technical problems that urgently need to be solved in the field of mine slope monitoring and early warning. Summary of the Invention

[0008] This invention addresses the problems in existing technologies, such as how to automatically distinguish between construction vibrations and actual landslides, how to automatically identify slope disaster types, and how to improve the early detection sensitivity of deep, hidden fractures. It provides a real-time mine risk early warning system based on the Internet of Things and artificial intelligence, thus solving the above problems.

[0009] This invention provides a real-time risk early warning system for mines based on the Internet of Things and artificial intelligence, comprising: Position sensors are pre-embedded in a fixed-distance array within the slope based on the slope's position and inclination, and are used to collect displacement signals at various depth points within the slope. Vibration sensors, which are installed in conjunction with position sensors at the same physical point inside the slope, are used to collect vibration response signals at the same point. The receiver is used to acquire the acquisition signals from the position sensor and vibration sensor in real time, and to combine the displacement acquisition signals and vibration acquisition signals at the same point into a coupled sensing data group according to the one-to-one correspondence between the position sensor and the vibration sensor. The timing module generates timing synchronization signals at a fixed frequency; The signal judgment module receives the timing synchronization signal, extracts the coupled sensing data groups of each corresponding node in the receiver, and extracts a six-dimensional feature vector for each point. The six-dimensional feature vector includes vibration amplitude level, vibration dominant frequency characteristics, cumulative displacement, displacement rate, depth distribution index, and temporal evolution index. The signal judgment module projects the six-dimensional feature vector onto a preset mode feature space, calculates the membership degree of the point relative to four modes: construction vibration mode, surface slip mode, structural failure mode, and safe mode, and determines the current mode based on the principle of maximum membership degree. The signal judgment module maintains a state transition probability matrix, tracks the transition direction of each point mode over time through the state transition probability matrix, and presets multiple evolution path templates. By matching the evolution path templates, the development stage of the slope disaster is determined. The integrated early warning module, based on the pattern judgment results of each point output by the signal judgment module, the pattern membership degree, the pattern transformation trend calculated by the state transition probability matrix, and the evolution path template matching results, combined with the pattern spatial distribution characteristics of multiple points within the slope section, determines whether the current overall working condition of the slope is simple construction vibration, surface landslide, or overall structural failure, and outputs early warning signals of the corresponding type and level.

[0010] In the present invention, a real-time mine risk early warning system based on the Internet of Things and artificial intelligence, as a preferred embodiment, describes the following behavior of the position sensor and vibration sensor under construction vibration mode: The displacement of the position sensor increases instantaneously to the peak value. After the vibration ends, the displacement rebounds and recovers to near the reference value. The displacement rebound rate is greater than 90%, meaning that more than 90% of the peak displacement of the vibration rebounds and recovers after the vibration ends. Moreover, the amplitude of the shallow displacement is significantly greater than that of the deep displacement, and the ratio of the peak displacement of the shallow layer to that of the deep layer is greater than 3. The vibration sensor amplitude suddenly increases beyond the vibration threshold, the main vibration frequency falls within the main frequency range of construction vibration (i.e., the main frequency of blasting vibration is 20 to 80 Hz, the main frequency of mechanical operation vibration is 5 to 20 Hz), the vibration duration is short (i.e., blasting vibration lasts 0.1 to 5 seconds, mechanical operation vibration is periodic and intermittent), and the vibration energy decays rapidly with depth.

[0011] The present invention discloses a real-time mine risk early warning system based on the Internet of Things and artificial intelligence. As a preferred embodiment, the performance of the position sensor and vibration sensor under surface slip mode is as follows: The displacement of the position sensor continues to accumulate without rebound. The displacement rate is in the range of 0.1 to 10 mm per day. The displacement is mainly concentrated in the shallow layer of the slope, that is, the ratio of shallow displacement to deep displacement is greater than 5. The displacement below the slip surface decreases sharply and approaches zero, and the displacement rate changes at a constant speed or slowly accelerating. The vibration sensor has a micro-vibration event rate of less than 0.1 times per hour, the vibration amplitude is comparable to the background noise level, the vibration frequency does not rise abnormally and is consistent with the environmental vibration frequency, and there are no obvious micro-fracture signal characteristics.

[0012] The present invention discloses a real-time mine risk early warning system based on the Internet of Things and artificial intelligence. As a preferred embodiment, the performance of the position sensor and vibration sensor under structural failure modes are as follows: The displacement of the position sensor continues to accelerate, with a displacement rate greater than 10 mm per day and a rate of change greater than zero, i.e., the second derivative is positive. The ratio of shallow displacement to deep displacement is between 1 and 2, indicating that the deep displacement is significant, and the active displacement area gradually expands from the deep layer to the shallow layer. The vibration sensor's micro-vibration event rate is higher than 1 per hour and continues to increase. The dominant vibration frequency shows a decreasing trend from high frequency to low frequency, that is, the dominant frequency in the initial micro-fracture stage is 100 to 500 Hz, the dominant frequency in the crack propagation stage is 20 to 100 Hz, and the dominant frequency in the overall instability stage is lower than 10 Hz, and the micro-vibration b value continues to decrease.

[0013] The present invention discloses a real-time mine risk early warning system based on the Internet of Things and artificial intelligence. As a preferred embodiment, the position sensor and vibration sensor behave as follows in safety mode: The daily displacement change of the position sensor is less than 0.01 mm per day. The displacement fluctuates randomly around zero without any trend. The displacement at each depth is close to zero and has no significant gradient. Moreover, the displacement fluctuation is highly correlated with environmental factors such as rainfall and temperature. The background vibration level of the vibration sensor is less than 0.1 mm / s, which is the ambient noise level. The main vibration frequency is stable between 1 and 5 Hz, which is the natural environmental vibration. The micro-vibration event rate is close to zero, and there are no abnormal vibration signals.

[0014] The present invention discloses a real-time mine risk early warning system based on the Internet of Things and artificial intelligence. In a preferred embodiment, the signal judgment module includes a fuzzy boundary processing unit. This unit uses a Gaussian membership function to calculate the membership degree of each mode. The membership degree calculation formula is as follows: μ_i=exp(-Σw_j×(x_j-c_ij)² / (2σ_ij²)) Where x_j is the j-th feature parameter, c_ij is the j-th feature center value of the i-th pattern, σ_ij is the corresponding standard deviation, and w_j is the feature weight; when the difference between the maximum membership degree and the second maximum membership degree is greater than 0.3, it is directly determined to be the pattern corresponding to the maximum membership degree; when the difference is less than or equal to 0.3, it is determined to be a transitional state and marked as a state in transition between the patterns corresponding to the first two membership degrees.

[0015] The present invention discloses a real-time risk early warning system for mines based on the Internet of Things and artificial intelligence. In a preferred embodiment, when the signal judgment module determines that the state is transitional, the timing module automatically switches the sampling frequency from the reference frequency f0 to a high-frequency sampling frequency n×f0, where n is a frequency multiplier and n≥5, to track the direction of mode evolution with higher time resolution; when the mode is clearly determined, the timing module returns to the reference sampling frequency.

[0016] In the present invention, a real-time risk early warning system for mines based on the Internet of Things and artificial intelligence is preferably configured such that the state transition probability matrix P in the signal judgment module is a 4x4 matrix, in the following form: ; The rows and columns correspond to construction vibration mode A, surface slip mode B, structural failure mode C and safety mode D, respectively. The matrix element P_ij represents the probability of changing from mode i to mode j, and the sum of the elements in each row is equal to 1. The signal identification module uses a six-dimensional feature vector as the observation sequence and four modes as hidden states. It employs a Hidden Markov Model (HMM) and the Viterbi algorithm to solve for the optimal state sequence, achieving smooth mode transition recognition. The state transition probability matrix P is continuously updated using Bayesian updates during system operation. The update formula is as follows: P_ij_new=(N_ij+α_ij) / (Σ_kN_ik+Σ_kα_ik) Where N_ij is the number of observed transitions from pattern i to pattern j, and α_ij is the prior distribution parameter.

[0017] The present invention discloses a real-time mine risk early warning system based on the Internet of Things and artificial intelligence. As a preferred embodiment, the signal judgment module has four preset evolution path templates, each defined by a corresponding pattern sequence, as follows: The construction cycle path template features a pattern sequence that alternates periodically between a safety mode and a construction vibration mode. A template for rainfall-induced surface landslide path, the model sequence is a safe mode through a deceleration creep stage, a uniform creep stage to an accelerated creep stage; The template for the evolution path of deep fractures shows a pattern sequence from a safe mode through the micro-fracture incubation stage, the fracture propagation stage, to the overall instability stage. The construction-induced failure path template shows that the mode sequence is a safe mode that enters the micro-fracture incubation stage after construction vibration and gradually expands to the overall instability stage. The signal judgment module performs dynamic time warping matching between the actual mode sequence within the set time window and each template. The template with the highest matching degree is the current evolution path, and based on this, it predicts the most likely mode to be entered in the next stage and the remaining warning time.

[0018] The present invention provides a real-time mine risk early warning system based on the Internet of Things and artificial intelligence. As a preferred embodiment, the comprehensive early warning module is equipped with a spatial distribution discrimination unit. The spatial distribution discrimination unit constructs a pattern spatial distribution map based on the pattern judgment results of each point within the slope section. When the active vibration points are distributed in a planar pattern along the slope and the time difference of vibration start at each point is proportional to the distance from the blast source, it is determined to be simple construction vibration. When the active displacement points are distributed in strips along the shallow layer of the slope and the displacement below the slip surface decreases sharply, it is identified as a surface landslide. When the active displacement point coincides with the active vibration point and the active area gradually expands from the deep layer to the shallow layer, it is judged as overall structural failure. The signal judgment module is equipped with an environmental factor decoupling unit. The environmental factor decoupling unit establishes a rainfall-displacement response model and a temperature-displacement response model. It subtracts the environmental deformation components caused by rainfall and temperature from the measured displacement to obtain the real creep displacement components before performing mode judgment. The comprehensive early warning module is equipped with a confidence assessment unit. The confidence assessment unit calculates the confidence of the model judgment result based on four factors: feature consistency, data quality, spatial consistency, and historical stability. When the confidence is lower than 60%, it outputs a data insufficiency prompt and triggers manual review.

[0019] The beneficial effects of this invention are as follows: (1) The present invention adopts the same-point displacement-vibration coupling sensing and state combination discrimination method, which solves the problem that the existing technology cannot automatically distinguish between construction vibration and real landslide, realizes the automatic elimination of false alarms of engineering vibration, and reduces the false alarm rate of early warning by more than 40% compared with the traditional single-parameter threshold alarm.

[0020] (2) The present invention adopts a four-mode classification and spatial distribution verification method, which solves the problem that the existing monitoring system only outputs the risk level and cannot distinguish the disaster type. It realizes the automatic identification of three types of working conditions: construction vibration, surface landslide and overall structural damage, and upgrades the emergency response from a one-size-fits-all approach to classified measures.

[0021] (3) The present invention adopts a microseismic multi-parameter evolution analysis and deep and shallow layer synchronous verification method to solve the problem of delayed identification of deep hidden fractures. By using features such as microseismic event rate, main frequency reduction and b-value reduction, the rock mass fracture incubation process can be captured in advance. The overall instability warning advance is improved by several hours to several days compared with pure displacement monitoring.

[0022] (4) The present invention adopts the state transition probability matrix and evolution path template matching method to solve the problem that the mode judgment is easy to jump and the evolution trend cannot be tracked. It realizes the smooth identification of mode transition and the trend prediction of the next stage. The warning information is upgraded from static state description to dynamic evolution prediction.

[0023] (5) The present invention adopts Gaussian membership function and transition state recognition method to solve the problem of repeated jumps near the critical point of traditional threshold alarm. It achieves smooth transition of warning level through fuzzy boundary processing. The transition state automatically triggers high frequency sampling, taking into account both long-term low power consumption operation and high-precision monitoring during critical periods.

[0024] (6) The present invention adopts the slope slope adaptive layout and environmental factor decoupling method, which solves the problems of low monitoring efficiency of fixed spacing array and the data authenticity of rainfall and temperature interference, so that the sensor density matches the stress distribution gradient, and the monitoring accuracy is improved by more than 25% under the same number, while eliminating data distortion caused by environmental deformation. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a real-time risk warning system architecture for mines based on the Internet of Things and artificial intelligence.

[0026] Figure label: 1. Position sensor; 2. Vibration sensor; 3. Receiver; 4. Timing module; 5. Signal judgment module; 6. Integrated early warning module. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Example 1

[0028] like Figure 1 As shown, a real-time risk early warning system for mines based on the Internet of Things and artificial intelligence includes: Position sensor 1 is pre-embedded in a fixed-distance array within the slope according to the slope position and inclination, and is used to collect displacement signals at various depth points inside the slope. Vibration sensor 2, corresponding to position sensor 1, is installed at the same physical point inside the slope to collect vibration response signals at the same point. Receiver 3 is used to acquire the acquisition signals of position sensor 1 and vibration sensor 2 in real time, and combine the displacement acquisition signal and vibration acquisition signal at the same point into a coupled sensing data group according to the one-to-one correspondence between position sensor 1 and vibration sensor 2. Timing module 4 generates timing synchronization signals at a fixed frequency; The signal judgment module 5 receives the timing synchronization signal from the timing module 4, extracts the coupled sensing data groups of each corresponding node in the receiver 3, and extracts the six-dimensional feature vector of each point. The six-dimensional feature vector includes vibration amplitude level, vibration dominant frequency characteristics, cumulative displacement, displacement rate, depth distribution index, and timing evolution index. The signal judgment module 5 projects the six-dimensional feature vector onto the preset mode feature space, calculates the membership degree of the point relative to four modes: construction vibration mode, surface slip mode, structural failure mode, and safety mode, and determines the current mode according to the maximum membership degree principle. The signal judgment module 5 maintains a state transition probability matrix, tracks the transition direction of each point mode over time through the state transition probability matrix, and presets multiple evolution path templates. By matching the evolution path templates, the development stage of the slope disaster is determined. The integrated early warning module 6, based on the pattern determination results, pattern membership, pattern transition trend calculated by the state transition probability matrix, and evolution path template matching results of each point output by the signal judgment module 5, combined with the pattern spatial distribution characteristics of multiple points within the slope section, determines whether the current overall working condition of the slope is simple construction vibration, surface landslide, or overall structural failure, and outputs early warning signals of the corresponding type and level.

[0029] The behavior of position sensor 1 and vibration sensor 2 under construction vibration mode is as follows: The displacement of position sensor 1 increases instantaneously to the peak value. After the vibration ends, the displacement rebounds and recovers to near the reference value. The displacement rebound rate is greater than 90%, that is, more than 90% of the peak displacement of the vibration rebounds and recovers after the vibration ends. Moreover, the amplitude of the shallow displacement is significantly greater than the amplitude of the deep displacement, and the ratio of the peak displacement of the shallow layer to the peak displacement of the deep layer is greater than 3. The vibration amplitude of vibration sensor 2 suddenly increases beyond the vibration threshold. The main vibration frequency falls within the main frequency range of construction vibration, i.e., the main frequency of blasting vibration is 20 to 80 Hz and the main frequency of mechanical operation vibration is 5 to 20 Hz. The vibration duration is short, i.e., the blasting vibration lasts for 0.1 to 5 seconds and the mechanical operation vibration is periodic and intermittent. Moreover, the vibration energy decays rapidly with depth.

[0030] The behavior of position sensor 1 and vibration sensor 2 in surface slip mode is as follows: The displacement of position sensor 1 is continuous and without rebound. The displacement rate is in the range of 0.1 to 10 mm per day. The displacement is mainly concentrated in the shallow layer of the slope, that is, the ratio of shallow displacement to deep displacement is greater than 5. The displacement below the slip surface decreases sharply and approaches zero. The displacement rate changes at a constant speed or slowly accelerating. The vibration sensor 2 has a micro-vibration event rate of less than 0.1 times per hour, the vibration amplitude is comparable to the background noise level, the vibration frequency does not rise abnormally and is consistent with the environmental vibration frequency, and there are no obvious micro-fracture signal characteristics.

[0031] The behavior of position sensor 1 and vibration sensor 2 under structural failure modes are as follows: The displacement of position sensor 1 continues to accelerate, with a displacement rate greater than 10 mm per day and a rate of change greater than zero, i.e., the second derivative is positive. The ratio of shallow displacement to deep displacement is between 1 and 2, indicating that the deep displacement is significant, and the active displacement area gradually expands from the deep layer to the shallow layer. The vibration sensor 2 shows a micro-vibration event rate of more than 1 per hour and continues to increase. The vibration main frequency shows a frequency reduction trend from high frequency to low frequency, that is, the main frequency in the initial micro-fracture stage is 100 to 500 Hz, the main frequency in the crack propagation stage is 20 to 100 Hz, and the main frequency in the overall instability stage is below 10 Hz. In addition, the micro-vibration b value continues to decrease.

[0032] In safe mode, the position sensor 1 and vibration sensor 2 behave as follows: The daily displacement change of the position sensor is less than 0.01 mm. The displacement fluctuates randomly around zero without any trend. The displacement at each depth is close to zero and has no significant gradient. Moreover, the displacement fluctuation is highly correlated with environmental factors such as rainfall and temperature. Vibration sensor 2 has a background vibration level of less than 0.1 mm / s, which is the environmental noise level. The main vibration frequency is stable between 1 and 5 Hz, which is the natural environmental vibration. The micro-vibration event rate is close to zero, and there are no abnormal vibration signals.

[0033] Signal judgment module 5 is equipped with a fuzzy boundary processing unit. The fuzzy boundary processing unit uses a Gaussian membership function to calculate the membership degree of each mode. The membership degree calculation formula is as follows: μ_i=exp(-Σw_j×(x_j-c_ij)² / (2σ_ij²)) Where x_j is the j-th feature parameter, c_ij is the j-th feature center value of the i-th pattern, σ_ij is the corresponding standard deviation, and w_j is the feature weight; when the difference between the maximum membership degree and the second maximum membership degree is greater than 0.3, it is directly determined to be the pattern corresponding to the maximum membership degree; when the difference is less than or equal to 0.3, it is determined to be a transitional state and marked as a state in transition between the patterns corresponding to the first two membership degrees.

[0034] When the signal judgment module 5 determines that it is in a transitional state, the timing module 4 automatically switches the sampling frequency from the reference frequency f0 to the high-frequency sampling frequency n×f0, where n is the frequency multiplier and n≥5, to track the mode evolution direction with higher time resolution; when the mode is clearly determined, the timing module 4 returns to the reference sampling frequency.

[0035] 8. A real-time risk early warning system for mines based on the Internet of Things and artificial intelligence according to claim 1, characterized in that the state transition probability matrix P in the signal judgment module 5 is a 4-row, 4-column matrix, in the following form: ; The rows and columns correspond to construction vibration mode A, surface slip mode B, structural failure mode C and safety mode D, respectively. The matrix element P_ij represents the probability of changing from mode i to mode j, and the sum of the elements in each row is equal to 1. Signal identification module 5 uses a six-dimensional feature vector as the observation sequence and four modes as hidden states. It employs a Hidden Markov Model (HMM) and the Viterbi algorithm to solve for the optimal state sequence, achieving smooth mode transition recognition. The state transition probability matrix P is continuously updated using Bayesian updates during system operation. The update formula is: P_ij_new=(N_ij+α_ij) / (Σ_kN_ik+Σ_kα_ik) Where N_ij is the number of observed transitions from pattern i to pattern j, and α_ij is the prior distribution parameter.

[0036] The signal judgment module 5 has four preset evolution path templates, each defined by a corresponding pattern sequence, namely: The construction cycle path template features a pattern sequence that alternates periodically between a safety mode and a construction vibration mode. A template for rainfall-induced surface landslide path, the model sequence is a safe mode through a deceleration creep stage, a uniform creep stage to an accelerated creep stage; The template for the evolution path of deep fractures shows a pattern sequence from a safe mode through the micro-fracture incubation stage, the fracture propagation stage, to the overall instability stage. The construction-induced failure path template shows that the mode sequence is a safe mode that enters the micro-fracture incubation stage after construction vibration and gradually expands to the overall instability stage. The signal judgment module 5 performs dynamic time warping matching between the actual mode sequence within the set time window and each template. The template with the highest matching degree is the current evolution path, and based on this, it predicts the most likely mode to be entered in the next stage and the remaining warning time.

[0037] The integrated early warning module 6 is equipped with a spatial distribution discrimination unit, which constructs a pattern spatial distribution map based on the pattern discrimination results of each point within the slope section. When the active vibration points are distributed in a planar pattern along the slope and the time difference of vibration start at each point is proportional to the distance from the blast source, it is determined to be simple construction vibration. When the active displacement points are distributed in strips along the shallow layer of the slope and the displacement below the slip surface decreases sharply, it is identified as a surface landslide. When the active displacement point coincides with the active vibration point and the active area gradually expands from the deep layer to the shallow layer, it is judged as overall structural failure. Signal judgment module 5 is equipped with an environmental factor decoupling unit. The environmental factor decoupling unit establishes a rainfall-displacement response model and a temperature-displacement response model. It subtracts the environmental deformation components caused by rainfall and temperature from the measured displacement to obtain the real creep displacement components before performing mode judgment. The comprehensive early warning module 6 is equipped with a confidence assessment unit. The confidence assessment unit calculates the confidence of the model judgment result based on four factors: feature consistency, data quality, spatial consistency, and historical stability. When the confidence is lower than 60%, it outputs a data insufficiency prompt and triggers manual review.

[0038] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A real-time risk early warning system for mines based on the Internet of Things and artificial intelligence, characterized in that, include: Position sensor (1) is pre-embedded in a fixed-distance array within the slope according to the slope position and slope, and is used to collect displacement signals at various depth points inside the slope. Vibration sensor (2), corresponding to the position sensor (1), is installed at the same physical point inside the slope to collect vibration response signals at the same point; The receiver (3) is used to acquire the acquisition signals of the position sensor (1) and the vibration sensor (2) in real time, and to combine the displacement acquisition signal and vibration acquisition signal at the same point into a coupled sensing data group according to the one-to-one correspondence between the position sensor (1) and the vibration sensor (2). The timing module (4) generates timing synchronization signals at a fixed frequency; The signal judgment module (5) receives the timing synchronization signal from the timing module (4), extracts the coupled sensing data group of each corresponding node in the receiver (3), and extracts the six-dimensional feature vector of each point. The six-dimensional feature vector includes vibration amplitude level, vibration main frequency characteristics, displacement accumulation, displacement rate, depth distribution index and timing evolution index. The signal judgment module (5) projects the six-dimensional feature vector to the preset mode feature space, calculates the membership degree of the point relative to the four modes of construction vibration mode, surface slip mode, structural failure mode and safety mode, and determines the current mode according to the maximum membership degree principle. The signal judgment module (5) maintains a state transition probability matrix, tracks the transformation direction of each point mode over time through the state transition probability matrix, and presets multiple evolution path templates. The development stage of the slope disaster is determined by matching the evolution path templates. The comprehensive early warning module (6) determines the current overall working condition of the slope as simple construction vibration, surface landslide or overall structural damage based on the pattern determination results, pattern membership degree, pattern transformation trend calculated by the state transition probability matrix, and evolution path template matching results output by the signal judgment module (5) and the pattern spatial distribution characteristics of multiple points within the slope section. It then outputs early warning signals of the corresponding type and level.

2. The real-time risk early warning system for mines based on the Internet of Things and artificial intelligence as described in claim 1, characterized in that: The position sensor (1) and the vibration sensor (2) behave as follows under construction vibration mode: The displacement of the position sensor (1) increases instantaneously to reach the peak value. After the vibration ends, the displacement rebounds and recovers to near the reference value. The displacement rebound rate is greater than 90%, that is, more than 90% of the peak displacement of the vibration rebounds and recovers after the vibration ends. Moreover, the amplitude of the shallow displacement is significantly greater than the amplitude of the deep displacement, and the ratio of the peak displacement of the shallow layer to the peak displacement of the deep layer is greater than 3. The vibration sensor (2) suddenly increases the vibration amplitude beyond the vibration threshold, the vibration frequency falls within the construction vibration frequency range, that is, the blasting vibration frequency is 20 to 80 Hz, the mechanical operation vibration frequency is 5 to 20 Hz, the vibration duration is short, that is, the blasting vibration lasts for 0.1 to 5 seconds, the mechanical operation vibration is periodically intermittent, and the vibration energy decays rapidly with depth.

3. The real-time risk early warning system for mines based on the Internet of Things and artificial intelligence as described in claim 1, characterized in that: The position sensor (1) and the vibration sensor (2) behave as follows in the surface slip mode: The position sensor (1) continuously accumulates displacement without rebound, with a displacement rate ranging from 0.1 to 10 mm per day. The displacement is mainly concentrated in the shallow layer of the slope, i.e., the ratio of shallow displacement to deep displacement is greater than 5. The displacement below the slip surface decreases sharply and approaches zero, and the displacement rate changes at a constant speed or slowly. The vibration sensor (2) has a micro-vibration event rate of less than 0.1 times per hour, the vibration amplitude is comparable to the background noise level, the vibration main frequency does not rise abnormally and is consistent with the environmental vibration main frequency, and there are no obvious micro-fracture signal characteristics.

4. A real-time mine risk early warning system based on the Internet of Things and artificial intelligence as described in claim 1, characterized in that: The position sensor (1) and the vibration sensor (2) behave as follows under the structural failure mode: The displacement of the position sensor (1) continues to accelerate and grow, with a displacement rate greater than 10 mm per day and a rate of change greater than zero, i.e., the second derivative is positive. The ratio of shallow displacement to deep displacement is between 1 and 2, i.e., the deep displacement is significant, and the active displacement area gradually expands from the deep layer to the shallow layer. The vibration sensor (2) has a micro-vibration event rate of more than 1 per hour and continues to increase. The vibration main frequency shows a frequency reduction trend from high frequency to low frequency, that is, the main frequency in the initial micro-fracture stage is 100 to 500 Hz, the main frequency in the crack propagation stage is 20 to 100 Hz, the main frequency in the overall instability stage is less than 10 Hz, and the micro-vibration b value continues to decrease.

5. A real-time mine risk early warning system based on the Internet of Things and artificial intelligence according to claim 1, characterized in that: In safe mode, the position sensor (1) and the vibration sensor (2) behave as follows: The position sensor (1) has a daily displacement change of less than 0.01 mm. The displacement fluctuates randomly near zero without any trend change. The displacement at each depth is close to zero and has no significant gradient. The displacement fluctuation is highly correlated with environmental factors such as rainfall and temperature. The vibration sensor (2) has a background vibration level of less than 0.1 mm / s, which is the environmental noise level, and a vibration frequency that is stable between 1 and 5 Hz, which is the natural environmental vibration. The micro-vibration event rate is close to zero, and there is no abnormal vibration signal.

6. A real-time mine risk early warning system based on the Internet of Things and artificial intelligence according to claim 1, characterized in that: The signal judgment module (5) is equipped with a fuzzy boundary processing unit. The fuzzy boundary processing unit uses a Gaussian membership function to calculate the membership degree of each mode. The membership degree calculation formula is as follows: μ_i=exp(-Σw_j×(x_j-c_ij)² / (2σ_ij²)) Where x_j is the j-th feature parameter, c_ij is the j-th feature center value of the i-th pattern, σ_ij is the corresponding standard deviation, and w_j is the feature weight; when the difference between the maximum membership degree and the second maximum membership degree is greater than 0.3, it is directly determined to be the pattern corresponding to the maximum membership degree; when the difference is less than or equal to 0.3, it is determined to be a transitional state and marked as a state in transition between the patterns corresponding to the first two membership degrees.

7. A real-time mine risk early warning system based on the Internet of Things and artificial intelligence according to claim 6, characterized in that: When the signal judgment module (5) determines that it is in a transitional state, the timing module (4) automatically switches the sampling frequency from the reference frequency f0 to the high-frequency sampling frequency n×f0, where n is the frequency multiplier and n≥5, to track the mode evolution direction with higher time resolution; when the mode is clearly determined, the timing module (4) returns to the reference sampling frequency.

8. A real-time mine risk early warning system based on the Internet of Things and artificial intelligence according to claim 1, characterized in that, The state transition probability matrix P in the signal judgment module (5) is a 4-row, 4-column matrix, in the following form: ； The rows and columns correspond to construction vibration mode A, surface slip mode B, structural failure mode C and safety mode D, respectively. The matrix element P_ij represents the probability of changing from mode i to mode j, and the sum of the elements in each row is equal to 1. The signal judgment module (5) uses a six-dimensional feature vector as the observation sequence and four modes as the hidden states. It employs a hidden Markov model and the Viterbi algorithm to solve for the optimal state sequence, thereby achieving smooth recognition of mode transitions. The state transition probability matrix P is continuously updated through Bayesian updates during system operation. The update formula is as follows: P_ij_new=(N_ij+α_ij) / (Σ_kN_ik+Σ_kα_ik) Where N_ij is the number of observed transitions from pattern i to pattern j, and α_ij is the prior distribution parameter.

9. A real-time mine risk early warning system based on the Internet of Things and artificial intelligence according to claim 1, characterized in that: The signal judgment module (5) has four preset evolution path templates, each defined by a corresponding pattern sequence, namely: The construction cycle path template features a pattern sequence that alternates periodically between a safety mode and a construction vibration mode. A template for rainfall-induced surface landslide path, the model sequence is a safe mode through a deceleration creep stage, a uniform creep stage to an accelerated creep stage; The template for the evolution path of deep fractures shows a pattern sequence from a safe mode through the micro-fracture incubation stage, the fracture propagation stage, to the overall instability stage. The construction-induced failure path template shows that the mode sequence is a safe mode that enters the micro-fracture incubation stage after construction vibration and gradually expands to the overall instability stage. The signal judgment module (5) performs dynamic time warping matching between the actual mode sequence within the set time window and each template. The template with the highest matching degree is the current evolution path, and based on this, it predicts the most likely mode to be entered in the next stage and the remaining warning time.

10. A real-time mine risk early warning system based on the Internet of Things and artificial intelligence according to claim 1, characterized in that: The integrated early warning module (6) is equipped with a spatial distribution discrimination unit. The spatial distribution discrimination unit constructs a pattern spatial distribution map based on the pattern judgment results of each point in the slope section. When the active vibration points are distributed in a planar pattern along the slope and the time difference of vibration start at each point is proportional to the distance from the blast source, it is determined to be simple construction vibration. When the active displacement points are distributed in strips along the shallow layer of the slope and the displacement below the slip surface decreases sharply, it is identified as a surface landslide. When the active displacement point coincides with the active vibration point and the active area gradually expands from the deep layer to the shallow layer, it is judged as overall structural failure. The signal judgment module (5) is equipped with an environmental factor decoupling unit. The environmental factor decoupling unit establishes a rainfall-displacement response model and a temperature-displacement response model. It deducts the environmental deformation components caused by rainfall and temperature from the measured displacement to obtain the real creep displacement components before performing mode judgment. The comprehensive early warning module (6) is equipped with a confidence assessment unit. The confidence assessment unit calculates the confidence of the model judgment result based on four factors: feature consistency, data quality, spatial consistency and historical stability. When the confidence is lower than 60%, it outputs a data shortage prompt and triggers manual review.

Images