Intelligent early warning method, system and equipment for karst tunnel collapse based on micro-seismic multi-precursor characteristics

Abstract

The invention relates to the technical field of karst tunnel collapse early warning, in particular to a karst tunnel collapse intelligent early warning method, system and equipment based on micro-seismic multi-precursor characteristics. According to the technical scheme, the method comprises the steps of multi-scale micro-seismic monitoring network construction, multi-modal micro-seismic feature processing, quantum neural network early warning model construction and self-adaptive early warning and risk assessment. According to the invention, through omnibearing data acquisition of the multi-scale micro-seismic monitoring network, different depth sensors and an intelligent transmission mode, monitoring is accurate and stable, through combination of multi-modal feature extraction fusion, time domain, frequency domain and waveform multi-dimensional information mining, surrounding rock conditions are accurately judged, early warning analysis is enhanced by using quantum characteristics through the quantum neural network, and the early warning accuracy is improved. And self-adaptive threshold adjustment and multi-factor risk assessment are matched, accurate early warning is achieved, collapse is efficiently prevented, and tunnel construction safety is powerfully guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of karst tunnel collapse early warning, and in particular to an intelligent early warning method, system and equipment for karst tunnel collapse based on microseismic multi-precursor characteristics. Background Art

[0002] With the rapid development of tunnel engineering construction, landslides, as a common and serious geological disaster in tunnel construction, have always been a key issue in the engineering field. In karst tunnels, due to their special geological structures, such as caves and dissolution fissures, the stability of the surrounding rock is more complex and changeable, and the risk of landslides is further increased.

[0003] Traditional landslide early warning methods mostly rely on surrounding rock deformation monitoring. However, for hard and brittle surrounding rocks, the deformation before a landslide is often very small, which makes this early warning method based on deformation monitoring ineffective in many cases. The emergence of microseismic technology provides a new way for landslide early warning. Microseismic is a low-energy elastic wave or stress wave that accompanies the generation, expansion and penetration of rock cracks. By collecting microseismic signals through microseismic equipment, the entire process of potential landslides from microfracture to macroscopic instability can be tracked, thereby achieving early warning.

[0004] However, there are many technical difficulties in using microseismic multi-precursor characteristics for karst tunnel collapse warning. One of the outstanding problems is the effective processing and comprehensive analysis of microseismic multi-characteristic parameters. Microseismic signals contain rich information, such as multiple characteristic parameters such as source location, magnitude, frequency, waveform characteristics, etc. These parameters are interrelated and have a highly nonlinear relationship. The current analysis method is difficult to fully explore the intrinsic relationship between these parameters, and it is impossible to accurately extract key characteristic information closely related to collapse from complex microseismic data. For example, under different karst geological conditions, the change patterns of microseismic characteristic parameters are different. How to accurately identify these differences and establish an adaptive early warning model is a major challenge.

[0005] In addition, in actual engineering, the mechanical properties of tunnel surrounding rocks and construction disturbance factors will affect the microseismic characteristics, resulting in a large amount of noise interference in microseismic data, further increasing the difficulty of accurately extracting effective early warning information. Existing data processing technologies often have difficulty in retaining the integrity and authenticity of microseismic characteristic parameters while removing noise, resulting in a significant reduction in the accuracy of early warnings.

[0006] In summary, the present application proposes an intelligent early warning method, system and equipment for karst tunnel collapse based on microseismic multi-precursor characteristics. Summary of the invention

[0007] The purpose of the present invention is to address the problem of low accuracy of karst tunnel collapse warning in the background technology, and to propose an intelligent warning method, system and equipment for karst tunnel collapse based on microseismic multi-precursor characteristics.

[0008] In a first aspect, the present invention provides an intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics, comprising the following steps:

[0009] Construction of multi-scale microseismic monitoring network: Based on the geological conditions of the karst tunnel, a multi-scale monitoring network is constructed, which includes shallow, middle and deep sensors of different types and additional arrays at key locations. Data is transmitted through a wireless ad hoc network system that combines low-power Bluetooth and ZigBee hybrid communications.

[0010] Multimodal microseismic feature processing: For the collected microseismic signals, time domain feature analysis, short-time Fourier transform frequency domain analysis, Hilbert-Huang transform to construct a time-frequency joint feature matrix and waveform morphology feature mining are performed, and then a multimodal deep belief network is used to deeply integrate the time-frequency domain joint features and waveform morphology features;

[0011] Construction of quantum neural network early warning model: Use quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector, and construct a quantum neural network with a hidden layer structure of quantum entanglement gates and quantum rotation gates. Take the quantum feature vector as input and the landslide warning result as output, and use the quantum back propagation algorithm to train the model to achieve intelligent early warning function based on quantum computing;

[0012] Adaptive early warning and risk assessment: The Kalman filter algorithm is used to filter and predict the microseismic characteristic data and dynamically adjust the warning threshold accordingly. At the same time, the threshold is corrected through the fuzzy logic reasoning system based on the construction progress, support conditions and geological condition changes. The hierarchical analysis method is used to determine the weights of multiple factors such as microseismic characteristics, geological strength indicators, groundwater level changes, and construction blasting vibrations, and the fuzzy comprehensive evaluation method is used to comprehensively evaluate the risk of tunnel collapse.

[0013] Optionally, the multi-scale microseismic monitoring network is constructed, specifically comprising the following steps:

[0014] According to the geological prediction model of karst tunnels, the tunnel surrounding rock is divided into three monitoring levels: shallow, middle and deep. High-resolution piezoelectric sensors are used in the shallow layer, with a spacing of 3-5 meters and distributed in a ring around the tunnel wall.

[0015] Electromagnetic induction microseismic sensors are used in the middle layer, arranged in a plum blossom shape with a spacing of 5-8 meters; fiber grating microseismic sensors are arranged in the deep layer, with a monitoring point arranged every 10-15 meters along the tunnel axis;

[0016] At the same time, additional micro-acceleration sensor arrays are installed in key locations of densely developed karst areas and fault fracture zones;

[0017] Construct a wireless ad hoc network transmission system using low-power Bluetooth and ZigBee hybrid communication technology, and intelligently switch the communication mode to transmit sensor data to the data processing center according to the distance between the sensor node and the data processing center and the signal interference situation;

[0018] Shallow piezoelectric sensors have a sensitivity of up to S pz (Unit: V / (m / s 2 )), and the sensor spacing d1 is set to d1=3+0.2r (meters) according to the tunnel radius r and the surrounding rock characteristics;

[0019] The spacing of electromagnetic induction microseismic sensors in the middle layer is d2 = 5 + 0.3r (meters);

[0020] The spacing of deep fiber Bragg grating microseismic sensors is d3=10+0.5r (meter).

[0021] Optionally, the multimodal microseismic feature extraction and fusion specifically include the following steps: performing time domain feature analysis on the collected microseismic signals, calculating the peak amplitude, duration, rise time and pulse count parameters, using short-time Fourier transform to perform frequency domain analysis to obtain the spectrum diagram and extract frequency domain features such as the main frequency, bandwidth, and spectrum center of gravity, using Hilbert-Huang transform to decompose the microseismic signal to obtain the eigenmode function and calculate its instantaneous frequency and instantaneous amplitude to construct a time-frequency joint feature matrix, deeply analyzing the geometric feature parameters such as concavity, symmetry, and steepness of the microseismic waveform, and using a multimodal deep belief network to fuse the time-frequency domain joint features and waveform morphology features;

[0022] When analyzing the time domain characteristics, the peak amplitude A p =max|s(t)|, duration T d =t end -t start , where t start The microseismic signal exceeds the threshold A for the first time t The time determined based on the background noise, t end is the time when the microseismic signal last exceeded the threshold, and the rise time T r =t peak -t start , where t peak is the time when the microseismic signal reaches its peak value, pulse count:

[0023]

[0024] Among them, p(i) is the function for judging the microseismic signal pulse. When the signal slope change rate exceeds k pWhen p(i) = 1, otherwise p(i) = 0, n is the number of signal sampling points;

[0025] Short-time Fourier transform: Let the window function be w(t) and the window length be T w ,but:

[0026]

[0027] Where t is time and f is frequency;

[0028] Main frequency:

[0029]

[0030] Bandwidth:

[0031] B w =f max- f min

[0032] Among them, f max and f min The energy in the spectrum is greater than a certain threshold E f The maximum and minimum frequencies, and the center of gravity of the spectrum, are determined based on the signal energy distribution:

[0033]

[0034] Optionally, in the multi-modal microseismic feature extraction and fusion, the Hilbert-Huang transform obtains the eigenmode function, the instantaneous frequency:

[0035]

[0036] Among them, c i (t) is the eigenmode function;

[0037] Instantaneous amplitude:

[0038]

[0039] When mining waveform features, the concave and convex The positive or negative value of determines the convexity of the waveform. If C>0, it is concave, and C<0 is convex. Symmetry:

[0040]

[0041] Among them, t0 is the midpoint time of the waveform, T is the half waveform time width, and the steepness The steepness is evaluated by the rate of change of K, and the slope change of K near the peak is calculated;

[0042] The multimodal deep belief network is composed of multiple stacked restricted Boltzmann machines. The number of visible layer units of the l-th layer RBM is The number of hidden layer units is The visible layer state vector is The hidden layer state vector is The energy function is:

[0043]

[0044] in, is the connection weight, is the visible layer bias, is the hidden layer bias;

[0045] Parameter update formula:

[0046]

[0047] Among them, ∈ is the learning rate.

[0048] Optionally, the construction of an intelligent early warning model based on a quantum neural network specifically includes the following steps: using quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector, constructing a quantum neural network, whose input layer is the microseismic feature vector after quantum encoding, and whose hidden layer uses quantum entanglement gates and quantum rotation gates to construct a complex quantum computing structure, and whose output layer is the landslide warning result, and uses a quantum back propagation algorithm to train the quantum neural network;

[0049] In the construction of the intelligent early warning model based on quantum neural network, the quantum bit encoding formula is:

[0050]

[0051] Among them, ψ i =arctan(x i ), |ψ> is the quantum state, x i It is an element in the microseismic feature vector. Through this encoding method, the microseismic feature vector is mapped to the quantum state space, and the superposition and entanglement of quantum states are used to explore the potential correlation between microseismic features;

[0052] In the construction of the intelligent early warning model based on quantum neural network, the quantum revolving door used by the quantum neural network is:

[0053]

[0054] The loss function uses mean square error:

[0055]

[0056] Where N is the number of samples, is the predicted output, y i is the actual output.

[0057] Optionally, the adaptive dynamic adjustment of warning threshold and risk assessment specifically include the following steps: using Kalman filtering algorithm to filter and predict microseismic characteristic data in real time, dynamically adjusting the warning threshold according to the change trend of the predicted value and the uncertainty estimation, and correcting the warning threshold through the fuzzy logic reasoning system in combination with the construction progress, support conditions and geological condition change factors in the tunnel; constructing a multi-factor karst tunnel collapse risk assessment system, using the hierarchical analysis method to determine the weights of microseismic characteristics, geological strength indicators of tunnel surrounding rocks, groundwater level changes, and construction blasting vibration factors, and using the fuzzy comprehensive evaluation method to comprehensively assess the collapse risk of the tunnel;

[0058] In the adaptive warning threshold dynamic adjustment and risk assessment, the Kalman filter algorithm assumes that the microseismic characteristic data vector is The state vector is The measurement matrix is ​​H, the state transfer matrix is ​​F, the process noise covariance matrix is ​​Q, the measurement noise covariance matrix is ​​R, and the prediction steps are: Update steps: When the prediction error covariance matrix P k When the determinant value is less than D0, the warning threshold Where α is the adjustment coefficient, is the change in predicted value.

[0059] Optionally, in the adaptive warning threshold dynamic adjustment and risk assessment, when the weight is determined by the hierarchical analysis method, the judgment matrix A is set to be (a ij ), where a ij To express the importance of factor i relative to factor j, first calculate the product of each row of A:

[0060]

[0061] Calculate M again i The nth root of Last pair After normalization, we get

[0062] In the adaptive warning threshold dynamic adjustment and risk assessment, the fuzzy comprehensive evaluation method {V1, V2, ... V m}, let the evaluation set be The single factor fuzzy evaluation matrix is: Fuzzy comprehensive evaluation results, where is the fuzzy composition operator, Take the greater or lesser operator or the weighted average operator.

[0063] In a second aspect, the present invention provides an intelligent early warning system for karst tunnel collapse based on microseismic multi-precursor characteristics, comprising:

[0064] Multi-scale microseismic monitoring module: It is composed of a variety of microseismic sensors arranged in layers according to the geological prediction model of the karst tunnel. The shallow layer is equipped with high-resolution piezoelectric sensors, which are spaced 3-5 meters apart and distributed in a ring around the tunnel wall; the middle layer is equipped with electromagnetic induction microseismic sensors, which are arranged in a plum blossom shape and spaced 5-8 meters apart; the deep layer is equipped with fiber grating microseismic sensors, with a monitoring point arranged every 10-15 meters along the axial direction of the tunnel, and micro acceleration sensor arrays are set in key parts of the karst dense development area and fault fracture zone; it also includes a wireless self-organizing network transmission submodule using low-power Bluetooth and ZigBee hybrid communication technology to transmit the data collected by the sensor to the data processing center;

[0065] Data processing and feature extraction module: used to process the collected microseismic signals, including time domain feature analysis of microseismic signals, obtaining peak amplitude, duration, rise time and pulse count, using short-time Fourier transform to perform frequency domain analysis to obtain frequency domain features such as main frequency, bandwidth, and spectrum center of gravity, using Hilbert-Huang transform to decompose microseismic signals to obtain eigenmode functions and calculate their instantaneous frequency and instantaneous amplitude to construct a time-frequency joint feature matrix, as well as analyzing geometric feature parameters such as concavity, symmetry, and steepness of microseismic waveforms, and fusing time-domain-frequency domain joint features and waveform morphology features through a multimodal deep belief network;

[0066] Intelligent early warning model module: including a quantum coding unit and a quantum neural network unit. The quantum coding unit uses quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector. The quantum neural network unit uses the quantum-encoded microseismic feature vector as input, and its hidden layer uses quantum entanglement gates and quantum rotation gates to construct a complex quantum computing structure. The output layer outputs the landslide warning results, and the module uses the quantum back-propagation algorithm to train the quantum neural network.

[0067] Warning threshold adjustment and risk assessment module: It includes a Kalman filter submodule and a multi-factor risk assessment submodule. The Kalman filter submodule uses the Kalman filter algorithm to filter and predict the microseismic characteristic data in real time, and dynamically adjusts the warning threshold according to the changing trend and uncertainty estimation of the predicted value. At the same time, the warning threshold is corrected through the fuzzy logic reasoning system in combination with the construction progress, support conditions and geological conditions in the tunnel; the multi-factor risk assessment submodule constructs a risk assessment system including microseismic characteristics, geological strength indicators of the tunnel surrounding rock, groundwater level changes, and construction blasting vibration factors. The hierarchical analysis method is used to determine the weight of each factor, and the fuzzy comprehensive evaluation method is used to comprehensively evaluate the collapse risk of the tunnel.

[0068] In a third aspect, the present invention provides an intelligent early warning device for karst tunnel collapse based on microseismic multi-precursor characteristics, comprising:

[0069] A memory for storing instructions; wherein the instructions are used to implement the intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics as described in the first aspect;

[0070] A processor is used to execute instructions in the memory.

[0071] Compared with the prior art, the present invention has at least one of the following beneficial technical effects:

[0072] By arranging different types of sensors in the shallow, middle and deep layers of the tunnel surrounding rock, all-round monitoring of the surrounding rock conditions at different depths of the tunnel is achieved, improving the comprehensiveness and accuracy of monitoring;

[0073] The multi-dimensional features can fully reflect the characteristics of microseismic events and provide rich information for accurately judging the state of surrounding rock. The multi-modal deep belief network is used to fuse microseismic features of different modes, which can learn the internal connection and complex relationship between the features, avoid the limitations of single feature analysis, improve the accuracy of comprehensive judgment of microseismic events, and thus more accurately evaluate the stability of tunnel surrounding rock.

[0074] Using quantum bits to encode the fused microseismic multimodal features can more effectively represent microseismic features and improve the model's ability to analyze microseismic signals.

[0075] By performing real-time filtering and prediction of microseismic characteristic data based on the Kalman filter algorithm, the reliability of early warning is improved, and the risk of tunnel collapse can be comprehensively and objectively assessed.

[0076] The present invention collects data in all directions through a multi-scale microseismic monitoring network. Different depth sensors and intelligent transmission modes make monitoring accurate and stable. It combines multi-modal feature extraction and fusion, mines multi-dimensional information in the time domain, frequency domain and waveform, accurately judges the surrounding rock conditions, and uses quantum neural networks to enhance early warning analysis using quantum characteristics. It is then combined with adaptive threshold adjustment and multi-factor risk assessment to provide accurate early warnings, effectively prevent landslides, and effectively ensure the safety of tunnel construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 This is a flow chart of the intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics. DETAILED DESCRIPTION

[0078] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0079] like Figure 1As shown, the present invention proposes an intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics, including four steps: multi-scale microseismic monitoring network construction, multi-modal microseismic feature processing, quantum neural network early warning model construction, and adaptive early warning and risk assessment. Each step is described in detail below.

[0080] 1. Construction of a multi-scale microseismic monitoring network: Based on the geological conditions of the karst tunnel, a multi-scale monitoring network is constructed that includes different types of sensors in the shallow, middle and deep layers, and additional arrays are added at key locations. Data is transmitted through a wireless ad hoc network system that uses a hybrid communication of low-power Bluetooth and ZigBee.

[0081] The specific steps include:

[0082] According to the geological prediction model of karst tunnels, the tunnel surrounding rock is divided into three monitoring levels: shallow, middle and deep. High-resolution piezoelectric sensors are used in the shallow layer, with a spacing of 3-5 meters and distributed in a ring around the tunnel wall.

[0083] Electromagnetic induction microseismic sensors are used in the middle layer, arranged in a plum blossom shape with a spacing of 5-8 meters; fiber grating microseismic sensors are arranged in the deep layer, with a monitoring point arranged every 10-15 meters along the tunnel axis;

[0084] At the same time, additional micro-acceleration sensor arrays are installed in key locations of densely developed karst areas and fault fracture zones;

[0085] Construct a wireless ad hoc network transmission system using low-power Bluetooth and ZigBee hybrid communication technology, and intelligently switch the communication mode to transmit sensor data to the data processing center according to the distance between the sensor node and the data processing center and the signal interference situation;

[0086] Shallow piezoelectric sensors have a sensitivity of up to S pz (Unit: V / (m / s 2 )), and the sensor spacing d1 is set to d1=3+0.2r (meters) according to the tunnel radius r and the surrounding rock characteristics;

[0087] The spacing of electromagnetic induction microseismic sensors in the middle layer is d2 = 5 + 0.3r (meters);

[0088] The spacing of deep fiber Bragg grating microseismic sensors is d3=10+0.5r (meter).

[0089] By arranging different types of sensors in the shallow, middle and deep layers of the tunnel surrounding rock, all-round monitoring of the surrounding rock conditions at different depths of the tunnel is achieved. The piezoelectric sensors in the shallow layer can keenly capture the changes in tiny rock cracks caused by construction disturbances and shallow karst caves; the electromagnetic induction sensors in the middle layer can monitor the stress changes of the surrounding rock over a larger range; and the fiber grating sensors in the deep layer focus on the changes in the deep rock structure.

[0090] By adding arrays of miniature accelerometers in key areas of densely developed karst areas and fault fracture zones, we can supplement the monitoring of microseismic signals under complex geological conditions, further improve the comprehensiveness and accuracy of monitoring, and not miss any potential hidden dangers that may cause landslides.

[0091] The wireless ad hoc network transmission system using low-power Bluetooth and ZigBee hybrid communication technology can intelligently switch the communication mode according to the signal interference and transmission distance in the tunnel. This not only ensures that the sensor data can be stably and efficiently transmitted to the data processing center, but also reduces power consumption, extends the service life of the sensor, and ensures long-term and reliable acquisition of monitoring data.

[0092] 2. Multimodal microseismic feature processing: For the collected microseismic signals, time domain feature analysis, short-time Fourier transform frequency domain analysis, Hilbert-Huang transform to construct a time-frequency joint feature matrix and waveform morphology feature mining are performed, and then a multimodal deep belief network is used to deeply fuse the time-domain-frequency domain joint features and waveform morphology features; Multimodal microseismic feature extraction and fusion specifically include the following steps: time domain feature analysis of the collected microseismic signals, calculation of peak amplitude, duration, rise time and pulse count parameters, use of short-time Fourier transform to perform frequency domain analysis to obtain the spectrum diagram and extract frequency domain features such as main frequency, bandwidth, and spectrum center of gravity, use Hilbert-Huang transform to decompose the microseismic signal to obtain the intrinsic mode function and calculate its instantaneous frequency and instantaneous amplitude to construct a time-frequency joint feature matrix, in-depth analysis of the geometric feature parameters of the microseismic waveform such as concavity, symmetry, and steepness, and use of a multimodal deep belief network to fuse the time-frequency domain joint features and waveform morphology features;

[0093] When analyzing the time domain characteristics, the peak amplitude A p =max|s(t)|, duration T d =t end -t start , where t start The microseismic signal exceeds the threshold A for the first time t The time determined based on the background noise, t end is the time when the microseismic signal last exceeded the threshold, and the rise time T r =t peak -t start , where t peak is the time when the microseismic signal reaches its peak value, pulse count:

[0094]

[0095] Among them, p(i) is the function for judging the microseismic signal pulse. When the signal slope change rate exceeds k pWhen p(i) = 1, otherwise p(i) = 0, n is the number of signal sampling points;

[0096] Short-time Fourier transform: Let the window function be w(t) and the window length be T w ,but:

[0097]

[0098] Where t is time and f is frequency;

[0099] Main frequency:

[0100]

[0101] Bandwidth:

[0102] B w =f max- f min

[0103] Among them, f max and f min The energy in the spectrum is greater than a certain threshold E f The maximum and minimum frequencies, and the center of gravity of the spectrum, are determined based on the signal energy distribution:

[0104]

[0105] Among them, the Hilbert-Huang transform obtains the eigenmode function, the instantaneous frequency:

[0106]

[0107] Among them, c i (t) is the eigenmode function;

[0108] Instantaneous amplitude:

[0109]

[0110] When mining waveform features, the concave and convex The positive or negative value of determines the convexity of the waveform. If C>0, it is concave, and C<0 is convex. Symmetry:

[0111]

[0112] Among them, t0 is the midpoint time of the waveform, T is the half waveform time width, and the steepness The steepness is evaluated by the rate of change of K, and the slope change of K near the peak is calculated;

[0113] The multimodal deep belief network is composed of multiple stacked restricted Boltzmann machines. The number of visible layer units of the l-th layer RBM is The number of hidden layer units is The visible layer state vector is The hidden layer state vector is The energy function is:

[0114]

[0115] in, is the connection weight, is the visible layer bias, is the hidden layer bias;

[0116] Parameter update formula:

[0117]

[0118] Among them, ∈ is the learning rate.

[0119] The microseismic signals are extracted from the joint time-domain and frequency-domain features, including the calculation of time-domain features such as peak amplitude, duration, rise time, pulse count, etc., and the use of short-time Fourier transform and Hilbert-Huang transform to obtain frequency-domain features and joint time-frequency feature matrices. At the same time, the waveform morphology characteristics are deeply explored, such as concavity, symmetry, and steepness. These multi-dimensional features can fully reflect the characteristics of microseismic events and provide rich information for accurately judging the state of the surrounding rock.

[0120] The use of a multimodal deep belief network to fuse microseismic features of different modes can learn the intrinsic connections and complex relationships between the features, avoid the limitations of single feature analysis, improve the accuracy of comprehensive judgment of microseismic events, and thus more accurately evaluate the stability of the tunnel surrounding rock.

[0121] 3. Construction of quantum neural network early warning model: Use quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector, construct a quantum neural network with a hidden layer structure of quantum entanglement gates and quantum rotation gates, take the quantum feature vector as input, and the landslide warning result as output, and use the quantum back propagation algorithm to train the model to achieve an intelligent early warning function based on quantum computing; The construction of an intelligent early warning model based on quantum neural network specifically includes the following steps: Use quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector, construct a quantum neural network, whose input layer is the microseismic feature vector after quantum encoding, and the hidden layer uses quantum entanglement gates and quantum rotation gates to construct a complex quantum computing structure, and the output layer is the landslide warning result, and the quantum back propagation algorithm is used to train the quantum neural network;

[0122] In the construction of the intelligent early warning model based on quantum neural network, the quantum bit encoding formula is:

[0123]

[0124] Among them, ψ i =arctan(x i ), |ψ> is the quantum state, x i It is an element in the microseismic feature vector. Through this encoding method, the microseismic feature vector is mapped to the quantum state space, and the superposition and entanglement of quantum states are used to explore the potential correlation between microseismic features;

[0125] In the construction of intelligent early warning model based on quantum neural network, quantum revolving door used by quantum neural network:

[0126]

[0127] The loss function uses mean square error:

[0128]

[0129] Where N is the number of samples, is the predicted output, y i is the actual output.

[0130] Quantum bits are used to encode the fused microseismic multimodal features. With the help of the superposition and entanglement of quantum states, deeper potential correlations and complex relationships between microseismic features can be excavated. Compared with traditional feature encoding methods, it can more effectively represent microseismic features and enhance the model's ability to analyze microseismic signals.

[0131] The construction and training of quantum neural networks, using the parallel computing power of quantum states, can accelerate the training speed of the model, enabling it to adapt to new data and environmental changes more quickly. At the same time, the quantum entanglement mechanism helps enhance the model's ability to learn the nonlinear relationship between microseismic characteristics, thereby more accurately predicting landslide risks.

[0132] The visual interpretation of the model output results through quantum state tomography technology helps engineering personnel understand the decision-making process of the early warning model, enabling them to take more targeted preventive measures and improve the scientific nature of construction decisions.

[0133] 4. Adaptive early warning and risk assessment: Use the Kalman filter algorithm to filter and predict the microseismic characteristic data and dynamically adjust the early warning threshold accordingly. At the same time, the threshold is corrected through the fuzzy logic reasoning system based on the construction progress, support conditions and geological condition changes. The hierarchical analysis method is used to determine the weights of multiple factors such as microseismic characteristics, geological strength indicators, groundwater level changes, and construction blasting vibrations. The fuzzy comprehensive evaluation method is used to comprehensively assess the risk of tunnel collapse.

[0134] The adaptive dynamic adjustment of warning threshold and risk assessment specifically include the following steps: using Kalman filter algorithm to filter and predict microseismic characteristic data in real time, dynamically adjusting the warning threshold according to the change trend and uncertainty estimation of the predicted value, and correcting the warning threshold through fuzzy logic reasoning system in combination with the construction progress, support conditions and geological condition changes in the tunnel; constructing a multi-factor karst tunnel collapse risk assessment system, using hierarchical analysis method to determine the weights of microseismic characteristics, geological strength indicators of tunnel surrounding rocks, groundwater level changes, and construction blasting vibration factors, and using fuzzy comprehensive evaluation method to comprehensively assess the collapse risk of the tunnel;

[0135] In the adaptive warning threshold dynamic adjustment and risk assessment, the Kalman filter algorithm is used to set the microseismic characteristic data vector as The state vector is The measurement matrix is ​​H, the state transfer matrix is ​​F, the process noise covariance matrix is ​​Q, the measurement noise covariance matrix is ​​R, and the prediction steps are: Update steps: When the prediction error covariance matrix P k When the determinant value is less than D0, the warning threshold Where α is the adjustment coefficient, is the change in predicted value.

[0136] In the adaptive warning threshold dynamic adjustment and risk assessment, when the analytic hierarchy process determines the weight, the judgment matrix A = (a ij ), where a ij To express the importance of factor i relative to factor j, first calculate the product of each row of A:

[0137]

[0138] Calculate M again i The nth root of Last pair After normalization, we get

[0139] In the adaptive warning threshold dynamic adjustment and risk assessment, the fuzzy comprehensive evaluation method {V1, V2, ... V m}, let the evaluation set be The single factor fuzzy evaluation matrix is: Fuzzy comprehensive evaluation results, where ° is the fuzzy synthesis operator, Take the greater or lesser operator or the weighted average operator.

[0140] Based on the Kalman filter algorithm, the microseismic characteristic data is filtered and predicted in real time, and the warning threshold is dynamically adjusted according to the change trend and uncertainty of the predicted value. This adaptive threshold adjustment mechanism can issue warning signals in a timely and accurate manner according to the actual microseismic activity in the tunnel, avoiding the problem of false alarms or missed alarms that may be caused by fixed thresholds, and improving the reliability of warnings.

[0141] By constructing a multi-factor karst tunnel collapse risk assessment system, in addition to microseismic characteristics, the geological strength index of the tunnel surrounding rock, groundwater level changes, and construction blasting vibration factors are also comprehensively considered. The weight of each factor is determined by the hierarchical analysis method, and the fuzzy comprehensive evaluation method is used for comprehensive assessment. It can comprehensively and objectively assess the tunnel collapse risk and provide detailed and accurate risk information for construction personnel, so that they can take reasonable construction adjustment measures according to the risk level to ensure the safety of tunnel construction.

[0142] In addition, the present invention provides an intelligent early warning system for karst tunnel collapse based on microseismic multi-precursor characteristics, which includes a multi-scale microseismic monitoring module, a data processing and feature extraction module, an intelligent early warning model module, and an early warning threshold adjustment and risk assessment module. Specifically:

[0143] Multi-scale microseismic monitoring module: It is composed of a variety of microseismic sensors arranged in layers according to the geological prediction model of the karst tunnel. The shallow layer is equipped with high-resolution piezoelectric sensors, which are spaced 3-5 meters apart and distributed in a ring around the tunnel wall; the middle layer is equipped with electromagnetic induction microseismic sensors, which are arranged in a plum blossom shape and spaced 5-8 meters apart; the deep layer is equipped with fiber grating microseismic sensors, with a monitoring point arranged every 10-15 meters along the axial direction of the tunnel, and micro acceleration sensor arrays are set in key parts of the karst dense development area and fault fracture zone; it also includes a wireless self-organizing network transmission submodule using low-power Bluetooth and ZigBee hybrid communication technology to transmit the data collected by the sensor to the data processing center;

[0144] Data processing and feature extraction module: used to process the collected microseismic signals, including time domain feature analysis of microseismic signals, obtaining peak amplitude, duration, rise time and pulse count, using short-time Fourier transform to perform frequency domain analysis to obtain frequency domain features such as main frequency, bandwidth, and spectrum center of gravity, using Hilbert-Huang transform to decompose microseismic signals to obtain eigenmode functions and calculate their instantaneous frequency and instantaneous amplitude to construct a time-frequency joint feature matrix, as well as analyzing geometric feature parameters such as concavity, symmetry, and steepness of microseismic waveforms, and fusing time-domain-frequency domain joint features and waveform morphology features through a multimodal deep belief network;

[0145] Intelligent early warning model module: including quantum coding unit and quantum neural network unit. The quantum coding unit uses quantum bits to encode the fused microseismic multimodal features to construct quantum feature vectors. The quantum neural network unit uses the quantum-encoded microseismic feature vectors as input. Its hidden layer uses quantum entanglement gates and quantum rotation gates to construct a complex quantum computing structure. The output layer outputs the landslide warning results. The module uses the quantum back-propagation algorithm to train the quantum neural network.

[0146] Warning threshold adjustment and risk assessment module: It includes a Kalman filter submodule and a multi-factor risk assessment submodule. The Kalman filter submodule uses the Kalman filter algorithm to filter and predict the microseismic characteristic data in real time, and dynamically adjusts the warning threshold according to the changing trend and uncertainty estimation of the predicted value. At the same time, the warning threshold is corrected through the fuzzy logic reasoning system based on the construction progress, support conditions and geological conditions in the tunnel; the multi-factor risk assessment submodule constructs a risk assessment system including microseismic characteristics, geological strength indicators of the tunnel surrounding rock, groundwater level changes, and construction blasting vibration factors. The hierarchical analysis method is used to determine the weight of each factor, and the fuzzy comprehensive evaluation method is used to comprehensively evaluate the collapse risk of the tunnel.

[0147] This embodiment provides an intelligent early warning device for karst tunnel collapse based on microseismic multi-precursor characteristics, including:

[0148] A memory for storing instructions; wherein the instructions are used to implement the above-mentioned intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics;

[0149] A processor is used to execute instructions in memory.

[0150] The above specific embodiments are only several optional embodiments of the present invention. Based on the technical solutions of the present invention and the relevant inspirations of the above embodiments, those skilled in the art can make various alternative improvements and combinations to the above specific embodiments.

Claims

1. An intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics, characterized in that: The following steps are involved: Construction of a multi-scale microseismic monitoring network: Based on the geological conditions of the karst tunnel, a multi-scale monitoring network including shallow, middle and deep sensors is constructed, and supplementary arrays are added at key locations. Data is transmitted through a wireless ad hoc network system that combines low-power Bluetooth and ZigBee hybrid communications; Multimodal microseismic feature processing: For the collected microseismic signals, time domain feature analysis, short-time Fourier transform frequency domain analysis, Hilbert-Huang transform to construct a time-frequency joint feature matrix and waveform morphology feature mining are performed, and then a multimodal deep belief network is used to deeply integrate the time-frequency domain joint features and waveform morphology features; Construction of quantum neural network early warning model: Use quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector, and construct a quantum neural network with a hidden layer structure of quantum entanglement gates and quantum rotation gates. Take the quantum feature vector as input and the landslide warning result as output, and use the quantum back propagation algorithm to train the model to achieve intelligent early warning function based on quantum computing; Adaptive early warning and risk assessment: The Kalman filter algorithm is used to filter and predict the microseismic characteristic data and dynamically adjust the warning threshold accordingly. At the same time, the threshold is corrected through the fuzzy logic reasoning system based on the construction progress, support conditions and geological condition changes. The hierarchical analysis method is used to determine the weights of multiple factors such as microseismic characteristics, geological strength indicators, groundwater level changes, and construction blasting vibrations, and the fuzzy comprehensive evaluation method is used to comprehensively evaluate the risk of tunnel collapse.

2. The intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics according to claim 1 is characterized in that: The multi-scale microseismic monitoring network construction specifically includes the following steps: According to the geological prediction model of karst tunnels, the tunnel surrounding rock is divided into three monitoring levels: shallow, middle and deep. High-resolution piezoelectric sensors are used in the shallow layer, with a spacing of 3-5 meters and distributed in a ring around the tunnel wall. Electromagnetic induction microseismic sensors are used in the middle layer, arranged in a plum blossom shape with a spacing of 5-8 meters; fiber grating microseismic sensors are arranged in the deep layer, with a monitoring point arranged every 10-15 meters along the tunnel axis; At the same time, additional micro-acceleration sensor arrays are installed in key locations of densely developed karst areas and fault fracture zones; Construct a wireless ad hoc network transmission system using low-power Bluetooth and ZigBee hybrid communication technology, and intelligently switch the communication mode to transmit sensor data to the data processing center according to the distance between the sensor node and the data processing center and the signal interference situation; Shallow piezoelectric sensors have a sensitivity of up to S pz , unit: V / (m / s 2 ), and the sensor spacing d1 is set to d1 = 3 + 0.2r (meters) according to the tunnel radius r and the surrounding rock characteristics; The spacing of electromagnetic induction microseismic sensors in the middle layer is d2 = 5 + 0.3r (meters); The spacing of deep fiber Bragg grating microseismic sensors is d3=10+0.5r (meter).

3. The intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics according to claim 1 is characterized in that: The multimodal microseismic feature extraction and fusion specifically includes the following steps: performing time domain feature analysis on the collected microseismic signals, calculating the peak amplitude, duration, rise time and pulse count parameters, using short-time Fourier transform to perform frequency domain analysis to obtain the spectrum diagram and extract frequency domain features such as the main frequency, bandwidth, and spectrum center of gravity, using Hilbert-Huang transform to decompose the microseismic signal to obtain the eigenmode function and calculate its instantaneous frequency and instantaneous amplitude to construct a time-frequency joint feature matrix, deeply analyzing the geometric feature parameters such as concavity, symmetry, and steepness of the microseismic waveform, and using a multimodal deep belief network to fuse the time-frequency domain joint features and waveform morphology features; When analyzing the time domain characteristics, the peak amplitude A p =max|s(t)|, duration T d =t end -t start , where t start The microseismic signal exceeds the threshold A for the first time t The time determined based on the background noise, t end is the time when the microseismic signal last exceeded the threshold, and the rise time T r =t peak -t start , where t peak is the time when the microseismic signal reaches its peak value, pulse count: Among them, p(i) is the function for judging the microseismic signal pulse. When the signal slope change rate exceeds k p When p(i) = 1, otherwise p(i) = 0, n is the number of signal sampling points; Short-time Fourier transform: Let the window function be w(t) and the window length be T w ,but: Where t is time and f is frequency; Main frequency: Bandwidth: B w =f max- f min Among them, f max and f min The energy in the spectrum is greater than a certain threshold E f The maximum and minimum frequencies, and the center of gravity of the spectrum, are determined based on the signal energy distribution:

4. The intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics according to claim 3 is characterized in that: In the multi-modal microseismic feature extraction and fusion, the Hilbert-Huang transform obtains the eigenmode function, the instantaneous frequency: Among them, c i (t) is the eigenmode function; Instantaneous amplitude: When mining waveform features, the concave and convex The positive or negative value of determines the convexity of the waveform. If C>0, it is concave, and C<0 is convex. Symmetry: Among them, t0 is the midpoint time of the waveform, T is the half waveform time width, and the steepness The steepness is evaluated by the rate of change of K, and the slope change of K near the peak is calculated; The multimodal deep belief network is composed of multiple stacked restricted Boltzmann machines. The number of visible layer units of the l-th layer RBM is The number of hidden layer units is The visible layer state vector is The hidden layer state vector is The energy function is: in, is the connection weight, is the visible layer bias, is the hidden layer bias; Parameter update formula: Among them, ∈ is the learning rate.

5. The intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics according to claim 1 is characterized in that: The construction of the intelligent early warning model based on quantum neural network specifically includes the following steps: using quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector, constructing a quantum neural network, whose input layer is the microseismic feature vector after quantum encoding, and the hidden layer uses quantum entanglement gates and quantum rotation gates to construct a complex quantum computing structure, and the output layer is the landslide warning result, and the quantum back propagation algorithm is used to train the quantum neural network; In the construction of the intelligent early warning model based on quantum neural network, the quantum bit encoding formula is: Among them, ψ i =arctan(x i ), |ψ> is the quantum state, x i It is an element in the microseismic feature vector. Through this encoding method, the microseismic feature vector is mapped to the quantum state space, and the superposition and entanglement of quantum states are used to explore the potential correlation between microseismic features; In the construction of the intelligent early warning model based on quantum neural network, the quantum revolving door used by the quantum neural network is: The loss function uses mean square error: Where N is the number of samples, is the predicted output, y i is the actual output.

6. The intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics according to claim 1 is characterized in that: The adaptive dynamic adjustment of warning threshold and risk assessment specifically include the following steps: using Kalman filtering algorithm to filter and predict microseismic characteristic data in real time, dynamically adjusting the warning threshold according to the change trend and uncertainty estimation of the predicted value, and correcting the warning threshold through a fuzzy logic reasoning system in combination with the construction progress, support conditions and geological condition changes in the tunnel; constructing a multi-factor karst tunnel collapse risk assessment system, using the hierarchical analysis method to determine the weights of microseismic characteristics, geological strength indicators of tunnel surrounding rocks, groundwater level changes, and construction blasting vibration factors, and using the fuzzy comprehensive evaluation method to comprehensively assess the collapse risk of the tunnel; In the adaptive warning threshold dynamic adjustment and risk assessment, the Kalman filter algorithm assumes that the microseismic characteristic data vector is The state vector is The measurement matrix is ​​H, the state transfer matrix is ​​F, the process noise covariance matrix is ​​Q, the measurement noise covariance matrix is ​​R, and the prediction steps are: Update steps: When the prediction error covariance matrix P k When the determinant value is less than D0, the warning threshold Where α is the adjustment coefficient, is the change in predicted value.

7. The intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics according to claim 6 is characterized in that: In the adaptive early warning threshold dynamic adjustment and risk assessment, when the analytic hierarchy process determines the weight, the judgment matrix A is set to be (a ij ), where a ij To express the importance of factor i relative to factor j, first calculate the product of each row of A: Calculate M again i The nth root of Last pair After normalization, we get In the adaptive warning threshold dynamic adjustment and risk assessment, the fuzzy comprehensive evaluation method {V1, V2, ... V m }, let the evaluation set be The single factor fuzzy evaluation matrix is: Fuzzy comprehensive evaluation results, where is the fuzzy composition operator, Take the greater or lesser operator or the weighted average operator.

8. An intelligent early warning system for karst tunnel collapse based on microseismic multi-precursor characteristics, characterized by: include: Multi-scale microseismic monitoring module: It is composed of a variety of microseismic sensors arranged in layers according to the geological prediction model of karst tunnels. The shallow layer is equipped with high-resolution piezoelectric sensors, which are spaced 3-5 meters apart and distributed in a ring around the tunnel wall; the middle layer is equipped with electromagnetic induction microseismic sensors, which are arranged in a plum blossom shape and spaced 5-8 meters apart; the deep layer is equipped with fiber grating microseismic sensors, with a monitoring point arranged every 10-15 meters along the axial direction of the tunnel, and micro acceleration sensor arrays are set in key parts of karst densely developed areas and fault fracture zones; it also includes a wireless self-organizing network transmission submodule using low-power Bluetooth and ZigBee hybrid communication technology to transmit the data collected by the sensor to the data processing center; Data processing and feature extraction module: used to process the collected microseismic signals, including time domain feature analysis of microseismic signals, obtaining peak amplitude, duration, rise time and pulse count, using short-time Fourier transform to perform frequency domain analysis to obtain frequency domain features such as main frequency, bandwidth, and spectrum center of gravity, using Hilbert-Huang transform to decompose microseismic signals to obtain eigenmode functions and calculate their instantaneous frequency and instantaneous amplitude to construct a time-frequency joint feature matrix, as well as analyzing geometric feature parameters such as concavity, symmetry, and steepness of microseismic waveforms, and fusing time-domain-frequency domain joint features and waveform morphology features through a multimodal deep belief network; Intelligent early warning model module: including a quantum coding unit and a quantum neural network unit. The quantum coding unit uses quantum bits to encode the fused microseismic multimodal features to construct a quantum feature vector. The quantum neural network unit uses the quantum-encoded microseismic feature vector as input, and its hidden layer uses quantum entanglement gates and quantum rotation gates to construct a complex quantum computing structure. The output layer outputs the landslide warning results, and the module uses the quantum back-propagation algorithm to train the quantum neural network. Warning threshold adjustment and risk assessment module: It includes a Kalman filter submodule and a multi-factor risk assessment submodule. The Kalman filter submodule uses the Kalman filter algorithm to filter and predict the microseismic characteristic data in real time, and dynamically adjusts the warning threshold according to the changing trend and uncertainty estimation of the predicted value. At the same time, the warning threshold is corrected through the fuzzy logic reasoning system in combination with the construction progress, support conditions and geological conditions in the tunnel; the multi-factor risk assessment submodule constructs a risk assessment system including microseismic characteristics, geological strength indicators of the tunnel surrounding rock, groundwater level changes, and construction blasting vibration factors. The hierarchical analysis method is used to determine the weight of each factor, and the fuzzy comprehensive evaluation method is used to comprehensively evaluate the collapse risk of the tunnel.

9. An intelligent early warning device for karst tunnel collapse based on microseismic multi-precursor characteristics, characterized in that: include: A memory for storing instructions; wherein the instructions are used to implement the intelligent early warning method for karst tunnel collapse based on microseismic multi-precursor characteristics according to any one of claims 1 to 7; A processor is used to execute instructions in the memory.

Images