CN111999765B - Micro-seismic multi-precursor method and device for early warning of tension-cracking falling type karst dangerous rock instability - Google Patents

Abstract

The invention discloses a micro-seismic multi-precursor method and device for early warning of fracture falling type karst dangerous rock instability, and aims to solve the problem of automatic early warning of fracture falling type dangerous rock collapse disasters in a karst area. Firstly, according to the microseismic change characteristics immediately before the instability of the dangerous rock, making 4 comprehensive relation rules of microseismic precursor characteristics and the stability grade of the dangerous rock; then, 4 microseismic precursor characteristics of each stage in the dangerous rock instability process are collected from two ways of indoor test and field example and are used as a machine learning sample set; and finally, establishing a LightGBM classification machine learning model with excellent adaptability to processing parallelization and large-scale high-dimensional data problems by utilizing a machine learning sample set, and establishing a nonlinear mapping relation between the microseismic precursor characteristics and the dangerous rock stability, so as to realize the rapid evaluation of the dangerous rock stability grade in online monitoring, and transmitting the evaluation result to an early warning information receiving terminal of a dangerous rock manager through an early warning device.

Description

Microseismic multi-precursor method and device for early warning of cracking and falling karst dangerous rock instability

technical field

The invention belongs to the technical field of geological disaster prevention engineering, and relates to a method and a device for early warning of cracking and falling karst dangerous rock instability and collapse by using microseismic signals.

Background technique

Dangerous rock refers to a geological body that is cut and separated by multiple groups of structural planes, has poor stability, and may collapse in the form of toppling, falling, and slipping. Cracked and falling karst dangerous rock refers to the karst area cut by fissures or the lower part is suspended, the dangerous rock mass on the steep slope, under the action of gravity and other factors, moves downward from the parent body, and finally accumulates at the foot of the slope, showing the mechanism of cracking failure. See attached picture 1.

The instability and collapse of dangerous rocks are highly sudden and destructive. The strong impact force directly causes the collapse and damage of the buildings below, which seriously affects the normal operation of road and railway traffic, not only causing huge property losses, Moreover, in recent years, there have been frequent incidents of casualties caused by dangerous rock collapses in tourist attractions. Dangerous rock collapse disasters mainly occur in Sichuan, Yunnan, Tibet, Guizhou, and Guangxi provinces (regions), with obvious regional characteristics, mainly reflected in the typical karst landform in these areas. Karst landform is also known as karst, and its rock mass is mainly composed of soluble The composition of the rock depends on the water, which is greatly affected by the water body, and has the characteristics of rich fissures and joints and poor stability. Moreover, the landform is widely distributed and covers a large area, and the dangerous rock instability and collapse disasters present the characteristics of various types, large scale, high incidence, wide distribution, and serious damage. In recent years, experts and scholars at home and abroad have conducted a lot of research on the cracking and falling type from different angles such as mechanical analysis, numerical calculation, and physical experiment, but the research progress is relatively slow, or the theoretical research results have been obtained in the karst crisis of the pulling and falling type. The accuracy and efficiency of rock stability grade identification are far from meeting the requirements, so that it is difficult to apply to practical engineering.

Microseism (Microseism, MS) (frequency < 100 Hz) is the stress concentration inside the rock mass under the influence of external disturbance stress and temperature, which causes the occurrence, expansion and penetration of microscopic cracks in the rock mass. Low energy elastic or stress waves. Microseismic signals contain rich information about the internal state changes of the rock mass, and can reflect the size, concentration, and rupture density of micro-cracks (large-scale) inside the rock mass.

The split and falling karst dangerous rock is one of the main types of single dangerous rock. Its stability is mainly controlled by the unloading and tensioning main control structural plane with a rear dip angle greater than 80°. The main control structural plane is located on the top of the dangerous rock mass, close to level and basically in a penetrating state. Under the combined action of gravity, earthquake force and fracture water pressure, the instability process presents the characteristics of tensile failure mechanics. In the process of instability, through rock micro-crack and micro-seismic monitoring and identification of instability precursor features, it is possible to effectively warn of dangerous rock instability, thereby avoiding the occurrence of dangerous rock collapse disasters.

The Light Gradient Boosting Machine (LightGBM) algorithm is a lightweight gradient boosting learning framework introduced by Microsoft. LightGBM is a typical representative algorithm under the idea of boosting reinforcement learning. Based on the XGboost algorithm, the algorithm uses the GOSS (gradient-based one-sided sampling) method and the EFB method (mutually exclusive feature bundling) to deal with the high memory in the Boost algorithm. The occupancy and processing performance are not powerful enough to further improve. Mainly based on the decision tree algorithm, it is an improvement to the gradient boosting decision tree (GBDT). The advantages are faster training efficiency, lower memory usage, higher accuracy, support for parallel learning and large-scale data processing, etc. It is widely used in various machine learning tasks such as sorting, classification, classification, etc., and has good promotion and application prospects and value.

The present invention introduces the LightGBM method into the pre-warning of the instability and collapse of the cracking and falling karst dangerous rock, and proposes a microseismic multi-premonition method and device for the early warning of the cracking and falling karst dangerous rock instability. The real-time monitoring and characteristic analysis of the microseismic signals during the collapse breeding process can realize the efficient and accurate identification of the stability of the cracked and fallen karst dangerous rocks, which has important practical value for the safety prevention and disaster prevention and mitigation of the cracked and fallen karst dangerous rocks.

Contents of the invention

The purpose of the present invention is to adopt rock mass rupture microseismic signal monitoring in view of the huge hazards of karst rock instability and collapse disasters caused by cracking and falling and the low reliability of existing early warning methods based on mechanical analysis, numerical calculation, and physical tests. Technical means, the light-weight gradient boosting tree machine learning method is introduced into the comprehensive early warning problem of cracking and falling karst dangerous rock instability and collapse based on various precursor characteristics of microseismic, and a method of early warning of cracking and falling karst dangerous rock instability is proposed. The microseismic multi-precursor method and device are used to effectively realize the reasonable early warning of the instability and collapse disaster of the karst dangerous rock that is pulled and dropped.

In order to achieve the above object, the present invention adopts the following technical solutions as follows:

On the one hand, the present invention provides a microseismic multi-precursor method for cracking and falling karst dangerous rock instability warning, including the following steps:

Step 1: According to the change characteristics and rules of the microseismic signal before the collapse of the karst dangerous rock instability and collapse, select a variety of significant precursor characteristic indicators of the dangerous rock instability as comprehensive early warning indicators, including: cumulative apparent volume, energy fractal dimension number, cumulative number of events, and b value; set the quantitative relationship between each precursory feature and the possibility of dangerous rock collapse instability, quantify various precursory characteristic indicators and the possibility of instability, and then formulate the The relationship rules between the microseismic precursor characteristic index and stability level of falling karst dangerous rock;

Step 2: Through extensive collection of indoor small rock sample tests and on-site project data of cracked and fallen karst dangerous rocks, four types of microseismic signal precursor features and corresponding stability levels are extracted to establish the original sample set for machine learning; according to the sample set, the The values of multiple precursor features under a certain stability level are combined to form a feature vector, which is used as an input vector of the model, and the corresponding stability level value is used as an output scalar of the LightGBM classification model, and an input A vector and an output scalar constitute a sample pair for training the LightGBM classification model. Similarly, the values of various precursory features under different stability levels are combined to form multiple feature vectors, and corresponding multiple stability level scalars form multiple training samples, thereby constructing a training sample set;

Step 3: Use the training sample set to train the LightGBM classification model, thereby constructing a nonlinear mapping relationship between multiple precursory feature indicators and stability levels;

Step 4: Using the trained LightGBM classification model, according to the characteristics of various microseismic precursors monitored in real time, carry out real-time identification of the stability level of the cracked and falling karst dangerous rock, and obtain the prediction result of the LightGBM classification model, that is, the stability level;

Step 5: Remotely transmit the early warning information to the dangerous rock manager.

Exemplarily, the present invention involves prediction and recognition of two keywords. It should be pointed out that the prediction mentioned in the present invention comes from the concept in the LightGBM classification model, and is not a prediction on a time scale; recognition refers to the Apply the LightGBM classification model to implement the identification of the stability level of the cracked and fallen karst dangerous rock mass; the two words appearing in the present invention are not confusing and conflicting, and can be understood as the form of prediction, and the purpose of identification.

Step 1 specific instructions:

Preferably, the cumulative apparent volume is the cumulative volume value of the rock mass in the inelastic deformation zone of the microseismic source, and the slope of its slope versus time curve is generally considered to be an important indicator of the strain rate of the rock mass, which can better reflect the microseismic breeding and development. The degree of damage to the dangerous rock mass during the process. Therefore, by accumulating the evolution law of apparent volume over time, the whole process of rock mass instability and collapse evolution can be better described. The evolution law of cumulative apparent volume is shown in Figure 2. a represents the initial static state, and the cumulative apparent volume value is less than 10 ³ /m ³ ; The apparent volume value rises faster, with a slope greater than 0.087; d indicates rapid sudden increase, and the apparent volume value presents a stepwise sudden increase. According to whether the overall trend is "initial static→slow rise→normal speed rise→rapid sudden increase", the whole process of rock mass instability and failure evolution can be warned. The present invention formulates the rules for the relationship between cumulative apparent volume and rock mass instability, as shown in Table 1.

Table 1 The relationship rules between cumulative apparent volume and rock mass instability possibility

Cumulative apparent volume initial static Slowly rising constant rate of ascent rapid burst Possibility of instability Small middle Big huge

Preferably, the energy fractal dimension is a measure of the change law of the energy of the described microseismic event. When the energy of the microseismic event changes little, the fractal dimension value is in a stable low value state; on the contrary, when the energy of the microseismic event changes significantly, the corresponding fractal dimension value rises rapidly. The energy fractal dimension of the rock mass increases with time before the macroscopic failure occurs. Specifically, the micro-cracks inside the rock mass continue to expand, the degree of damage and deterioration continues to increase, and at the same time, the accumulation of local damage is accelerated; and large-scale, high-energy microseismic events are gradually generated, and the release of microseismic energy further increases; when the accumulated microseismic release energy exceeds After the strain energy stored in a certain volume of the rock mass, macroscopic cracks appear and quickly gather in the weak area to form a certain scale of fracture surface. At this time, the energy fractal dimension reaches the maximum value, and then the rock mass becomes unstable. Therefore, the evolution of the fractal dimension of microseismic energy over time can better describe the whole process of rock mass instability and collapse evolution. The evolution law of the energy fractal dimension is shown in Figure 3. a indicates a steady fluctuation, and the value of the energy fractal dimension fluctuates below 0.2 or rises slowly with a slope of less than 0.157; b indicates a sudden increase point, and the value of the energy fractal dimension increases compared with the previous 20%; c means that the high value is stable, and the energy fractal dimension has multiple high value points compared with the initial energy fractal dimension value; d means that it rises rapidly, and the energy fractal dimension is in the high value stable stage, and its value is greater than A slope of 1 rises. The whole process of rock mass instability and failure evolution can be warned according to whether the overall trend is "stable fluctuation → sudden increase point → high value stable → rapid rise". The present invention formulates the relationship rules between energy fractal dimension and rock mass instability, as shown in Table 2.

Table 2 The relationship rules between the energy fractal dimension and the possibility of rock mass instability

energy fractal dimension smooth fluctuation burst point high value stable rise rapidly Possibility of instability Small middle Big huge

Preferably, the cumulative number of events refers to the cumulative value of the number of events in all time periods before this moment; the above-mentioned number of events refers to the number of times that the waveform of the microseismic signal passes through the set threshold within a frame, and it can to a certain extent It reflects the degree of activity of microseismic signals. High activity corresponds to a high number of events, and low activity corresponds to a low number of events. It can be seen that the number of events can reflect the process of each fracture stage of rock mass to a certain extent. The evolution law of the cumulative number of events is shown in Figure 4, a indicates a slow increase, the cumulative number of events remains at a low level below 500, and the slope is less than 0.123; b indicates a sudden increase point, and the cumulative event value has increased by more than 50% compared with the previous value; c indicates a rapid increase, and the cumulative number of events increases with a slope greater than 0.268; d indicates an accelerated increase, and the cumulative number of events increases with a slope greater than 1.428. According to the overall trend of the cumulative number of events, it shows "slow rise→sudden increase point→rapid rise→accelerated rise" to reflect the evolution process of rock mass instability. The present invention formulates the rules for the relationship between the cumulative number of events and the possibility of rock mass instability, as shown in Table 3.

Table 3 The relationship rules between the cumulative number of events and the possibility of rock mass instability

Cumulative number of events Slowly rising burst point rise rapidly speed up Possibility of instability Small middle Big huge

Preferably, the b value is an important parameter used to measure the level of seismic activity in a certain area, which is an important parameter for controlling the release capacity of accumulated energy by the medium. (1) When the rock mass is stable, the b value generally remains unchanged; (2) When micro-cracks are generated and developed inside the rock mass, the b value gradually increases due to the increase of small-magnitude micro-earthquakes; (3) when the rock mass approaches In the event of instability and collapse, the value of b drops sharply due to the increase of large-magnitude events and the decrease of small-magnitude events (see Figure 5). Therefore, it is possible to study the variation law of b value of microseismic earthquakes during the evolution of rock mass instability and collapse, to reveal the precursory characteristics of rock mass instability and collapse, and to serve as a basis for predicting rock mass instability and collapse. The evolution law of b value is shown in Figure 5. a represents low steady fluctuation, and b value fluctuates up and down in a low level interval; b represents rapid rise, and b value rises with a slope greater than 1; Fluctuates up and down in the high-level interval; d indicates a rapid decline, and the b value decreases with a high-level value and a slope less than -1. According to whether the overall trend shows "low steady fluctuation→rapid rise→high steady fluctuation→rapid decline", the whole process of rock mass instability and failure evolution can be warned. The present invention formulates the relationship rule between microseismic signal b value and rock mass instability, as shown in Table 4.

Table 4b The relationship between the value and the possibility of rock mass instability

b value low smooth fluctuation rise rapidly High smooth fluctuation drop rapidly Possibility of instability Small middle Big huge

The methods for obtaining the four indicators of precursory features are as follows:

The steps to obtain the cumulative apparent volume are as follows:

Step (1), calculate the microseismic radiation energy E:

where γ _eff is the effective surface energy, A is the fracture area with displacement u _i , Δσ _ij is the stress drop, n _j is the unit normal vector of the fracture surface, and σ _ij is the elongation rate. Radiation microseismic energy E is caused by the conversion of rock mass from elastic deformation to inelastic deformation during rock mass cracking and frictional slip.

Step (2), calculate the microseismic body potential P:

P = ΔεV (2)

其中A为震源面积，

是平均滑移量，P的量纲为[m·m₂]。它代表震源处发生非弹性变形区域的岩体体积的改变量，它与形状无关，可以从波形记录可靠算得。In the above formula, for a plane shear source, the microseismic body potential is defined as

where A is the source area,

is the average slip, and the dimension of P is [m·m ₂ ]. It represents the volume change of the rock mass in the region where the inelastic deformation occurs at the source. It has nothing to do with the shape and can be reliably calculated from the waveform record.

Step (3), calculate apparent volume V _A :

In the formula, E is the microseismic radiant energy; μ is the shear stiffness; P is the microseismic body change potential.

In step (4), the cumulative viewing volume ΣV _A is obtained.

The steps to obtain the energy fractal dimension are as follows:

Step (1), calculating the correlation integral of the energy distribution of the microseismic event produced by the microfracture of the rock mass:

In the formula: E is the upper limit value of the range interval of the microseismic release energy of all microseismic events; e is the microseismic release energy within the range of E; N is the logarithm value of the cumulative event number within the energy range of e; N is the microseismic event within the energy range of E total;

Step (2), with 1ge as the abscissa and lgc(e) as the ordinate, calculate the energy fractal dimension by establishing a rectangular coordinate system and performing linear fitting:

If the fitting line has a good linear correlation, it shows that the generation of rock micro-cracks has a fractal distribution relationship in energy.

The steps to obtain the cumulative number of events are as follows:

Step (1), process the microseismic signal by frame:

y _i (n)=w(n)*x((i-1)*inc+n) 1≤n≤L, 1≤i≤fn (6)

In the formula, ω(n) is a window function, generally a rectangular window or a Hamming window; y _i (n) is the value of a frame, n=1, 2,..., L, i=1, 2,..., f _n , L is the frame length; inc is the frame shift length; f _n is the total number of frames after framing. In this paper, a rectangular window is selected, and its function is as follows:

where the window length is L.

Step (2), calculate the number of events of the i-th frame microseismic signal y(n):

In the formula, sgn[y _i (n)] is a symbolic function, and its formula is as follows:

In the formula, the value of threshold is not fixed and should be determined according to specific application conditions.

Step (3), and then accumulating the number of events before the i frame, we can get:

The steps to obtain the value of b are as follows:

A large number of tests show that the microseismic events of rock mass instability and collapse all obey the magnitude-frequency (G-R) relational expression. By studying the microseismic activity, the magnitude-frequency relation of earthquakes is applicable to earthquakes in all magnitude ranges. The present invention passes linear Calculation of b-values of microseisms by least squares method:

In the formula, Δm is the bin interval of microseismic events, _Mi is the median number of microseismic events in bin i, and the b value is not only a function of the relative magnitude distribution of microseismic events, but also a function of the expansion scale of microseismic events.

Preferably, the above four kinds of precursor characteristic indicators based on microseismic signal rock mass instability and failure evolution can better describe the whole process of rock mass micro-destruction to macro-destruction and instability evolution. But, the sensitivity of each precursory feature is different for different rock mass types, and when it is possible that this kind of rock mass is critically damaged, a certain precursory feature does not appear or is not obvious; and, the application background of the present invention is in the nature In such a complex environment, the acquisition noise of each sampling point of the microseismic sensor changes, and it may be caused by the interference of noise when the cracked and falling karst dangerous rock collapses. In summary, a single microseismic signal with multiple precursor features describes the instability and collapse of karst dangerous rocks with a large randomness and weak anti-interference ability. The robustness of the comprehensive pre-warning of instability and collapse is low. Therefore, the present invention comprehensively considers the various precursory characteristics of the microseismic signals of the above-mentioned 4 kinds of cracking and falling karst dangerous rocks, and integrates them to analyze the full range of the cracking and falling karst dangerous rocks. This can improve the accuracy and low robustness of early warning of single microseismic signal precursor features, and further increase the length of early warning of cracked and fallen karst dangerous rocks.

Preferably, the present invention is based on a large number of experiments on the relationship between microseismic signals and karst rock sample failure evolution, microseismic signal research documents of hard and brittle rock mass at home and abroad, and engineering cases of tension and falling karst dangerous rock instability and collapse, and according to the results obtained in step 1 The developed microseismic precursor characteristics and rock mass instability and failure rules table comprehensively consider the various precursory characteristics of the microseismic signal of the cracking and falling karst dangerous rocks, and carry out the precursory characteristics of the cracking and falling karst dangerous rocks during the collapse evolution process. After analysis and quantification, a comprehensive rule table for its stability level was formulated, as shown in Table 5.

Table 5 Comprehensive relationship rules between microseismic precursory characteristics and stability levels of cracked and falling karst dangerous rocks

Step 2 specific instructions:

For step 2, including sub-steps 2.1, 2.2 and 2.3, the specific description is as follows.

Step 2.1: Microseismic signal preprocessing

Since the monitoring background of the present invention is a complex natural environment, it is easily disturbed by various types of factors such as climate, weather, environment, and construction. Defective data, in order to improve the overall quality of data.

The present invention discovers and corrects predictable errors in data files based on microseismic signals collected in real time from examples of cracked and fallen karst dangerous rock samples. The processing measures include: checking data consistency, and processing invalid and missing values.

Preferably, the measures include: missing value cleaning, determining its range, removing unnecessary fields, filling missing content, and re-fetching; logic error cleaning, removing duplicates, removing unreasonable values, and correcting contradictory content; non-demand data cleaning, Delete unnecessary redundant data; perform data cleaning according to the above sequence to obtain clean and optimized data.

Step 2.2: Microseismic signal precursor feature extraction

According to the microseismic signal precursor feature extraction method described in step 1, from the optimized microseismic signal of the cracked and falling karst dangerous rock, four features are proposed: cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value, and the precursory The feature quantization data x _i and the stability level to be predicted y _i form a sample ( _xi , y _i ).

Step 2.3: Create a sample

According to the multi-precursor feature data set of the microseismic signal of each sample of the cracked and fallen karst dangerous rock and its stability level, a machine learning sample ( _xi , y _i ) is established, where i=1,2,...,n, x _i is the input feature vector, where x _i =[ _xi1 , x _i2 , x _i3 , x _i4 ,], and each element is divided into four types of microseismic signals: cumulative apparent volume, energy fractal dimension, cumulative event number and b value A variety of precursory features; y _i is the output result of the stability grade of the cracked and fallen karst dangerous rock.

Step 3 specific instructions:

Preferably, the present invention adopts the basic idea of decomposing multi-category classification problems into multiple binary classification problems, and realizes the multi-class classification of the stability levels of cracked and fallen karst dangerous rocks by combining multiple LightGBM binary classification models. According to the "one-to-many" combination strategy, in order to realize the four categories of stability classification, it is necessary to establish and combine "good stability (Ⅰ)", "general stability (II)", "poor stability (Ⅲ) ", "Poor stability (Ⅳ)" four LightGBM binary classification models.

Lightweight Gradient Boosting Tree (LightGBM) is an algorithm of Gradient Boosting type. The LightGBM algorithm is based on the GBDT model and uses the gradient-based One-Side Sampling algorithm (Gradient-based One-SideSampling, GOSS) to preserve the gradient. At the same time, random sampling is used for data instances with small gradients, which realizes the perfect learning of under-trained samples and avoids excessive impact on the original distribution, increases the diversity of basic learners, and improves the generalization of the model Performance; at the same time, in the process of feature merging, a large number of features are combined into a smaller number and denser feature bundles by using the exclusive feature bundle algorithm (ExclusiveFeature Bundling, EFB), effectively avoiding the calculation of zero-valued features; combined with GOOS After the two optimization algorithms, the GBDT algorithm and the EFB algorithm, the serious shortcomings of the GBDT algorithm in the calculation cost have been improved, and the purpose of significantly reducing the time and space complexity of the algorithm has been achieved.

Step 3.1: Construction of LightGBM binary classification model

In the learning process of the LightGBM binary classification model, first, define and initialize the weak learner as f ₀ , which is the initial value of the model. Also define:

In the formula, x is the input sample, h is the t-th classification tree, ω is the parameter of the classification tree, and α is the weight of each tree in the prediction function. From the definition of this formula, it can be seen that the prediction function F is based on several An additive model formed by adding weak learners f ₀ .

相应的预测目标函数为

表达式为L(y_i,F(x_i))＝(y_i-F(x_i))²，为使预测损失函数减小得最快，其中

表示第i次迭代的弱学习器的建立方向，也称为响应：In each iteration, a weak learner based on classification tree is constructed, and the training samples are

The corresponding prediction objective function is

The expression is L(y _i ,F( _xi ))=(y _i -F( _xi )) ² , in order to reduce the prediction loss function the fastest, where

Denotes the building direction of the weak learner for the i-th iteration, also called the response:

即：Then, based on the obtained gradient descent direction, a decision tree is trained using the squared error to fit the data

which is:

Use the optimal step size to search in this direction, ie:

At this time, the value of the tth tree should be expressed as:

f _t = ρ ^* h _t (x; ω ^* ) (16)

Update the prediction function obtained after each iteration, namely:

F _t (x) = F _t-1 + f _t (17)

Repeat the above process continuously, and finally successfully solve the prediction function F, obtain the optimal model F*, that is, the prediction function that minimizes the loss function L, and make predictions for unknown samples:

Finally, for the binary classification problem involved in the present invention, after obtaining the above-mentioned final prediction function F*, the stability grade classification of the cracked and falling karst dangerous rock is realized, +1 means it belongs to this category, and -1 means it does not belong to this category .

In the training process of the LightBGM model, two important features of the GOSS algorithm and the EFB algorithm are involved.

对数据实例进行分割：First, the GOSS algorithm. Its purpose is to increase the utilization of data instances to a greater extent without affecting the distribution of data. The process is: sort the gradient size of each training data instance, which is divided into two cases; first, keep the top a×100% with a larger data instance, and use it as the data instance subset A; after that, for the remaining (1-a) × 100% data set A ^c of data instances with smaller gradients, a subset B is formed by random sampling, whose size is b × |A ^c |; finally, the variance gain on the set A ∪ B

Split a data instance:

用来将B的梯度和归一到Ac的大小，通过GOSS算法，用由一个更小子集计算得到的

代替根据所有数据实例计算得到的V_j(d)来判断分割点，同时，大大降低了计算成本。In the formula, A _r = {x|∈A: x _ij ＞d}, A _l = {x _i ∈ A: x _ij ≤ d}, B _r = {x _i ∈ B: x _ij ＞d}, B _l = {x _i ∈ B: x _ij ≤ d}, meanwhile, the coefficient

Used to normalize the gradient sum of B to the size of Ac, through the GOSS algorithm, calculated by a smaller subset

Instead of judging the segmentation point based on V _j (d) calculated from all data instances, at the same time, the calculation cost is greatly reduced.

Second, the EFB algorithm. Its purpose is to reduce the number of features of high-dimensional data in an approximately lossless manner, mainly from two aspects; on the one hand, it is necessary to identify important features that need to be merged; on the other hand, to merge mutually exclusive features. First, the feature is seen as the vertex of the graph, and the most characteristic binding problem is transformed into a graph coloring problem, so that a greedy algorithm that is not considered to be optimal overall but considered locally optimal in a certain sense can be used; second. , using the histogram algorithm (Histogram) to merge mutually exclusive features, the specific method is as follows: first, the continuous attribute values are discretized into m integers, and a histogram is constructed at the same time as the discretization, with a width of m; then, the data When traversing, the value after discretization can be used as the cumulative statistics of the index in the histogram; finally, it can be traversed according to the discrete value of the histogram to find the optimal segmentation point. This method effectively avoids huge and useless calculations, and further speeds up the performance of the LightGBM classification model.

In step 3.1, multiple generations of Light binary classification models are obtained through continuous iteration. In order to evaluate the optimal model, it is necessary to apply a measurement standard to judge. The model evaluation index used in the present invention is root mean square error RMSE (root mean squared error), the calculation formula is as follows:

In the formula, y _i* is the actual stability level value of the test sample, y _i is the predicted value of the LightGBM classification model for the comprehensive identification of the stability level of the cracked and falling karst dangerous rock with multiple precursors of microseismic, N is the sample size; the test index RMSE The LightGBM classification model corresponding to the minimum value is the optimal model, and the objective formula can be expressed as min{RMSE _i }, i=1,2,3,...,m.

Step 3.2: Feasibility test of LightGBM binary classification model

In order to ensure that the performance of the LightGBM classification model for the comprehensive identification of the stability level of the optimal microseismic multi-precursor cracking and falling karst dangerous rock meets the requirements (learning ability and generalization ability), the feasibility of the output test sample results of the optimal LightGBM classification model test. Specifically, the test index is the prediction accuracy rate of the test sample, that is, the actual stability level of the test sample and the predicted stability level are used to check. If the prediction accuracy rate is above 95%, it is considered that the performance of the optimal LightGBM classification model established meets the requirements. , it is feasible for the prediction of the stability level of karst dangerous rocks with cracking and falling; otherwise, retrain and build the model.

In step 3, the present invention adopts a typical k-fold cross validation (k-fold cross validation, K-CV) method, randomly divides the training sample library into 10 (k=10) parts, and selects 9 of them as training The other sample is used as a test sample, and the initial parameters such as num_leaves, learning_rate, max_depth and feature_fraction of the LightGBM classification model for the comprehensive identification of the stability level of the cracking and falling karst dangerous rock with multiple precursors are set, and the LightGBM classification model is used for learning and analysis. Prediction, and use the average learning accuracy rate and prediction accuracy rate of k calculations to evaluate the learning and generalization (extrapolation prediction) performance of the model.

In step 3, the present invention makes adjustments based on the cross-validation results of the LightGBM classification model for the comprehensive identification of the stability levels of cracked and fallen karst dangerous rocks with multiple precursors of microseismic events. If the performance of the cross-validated LightGBM classification model does not meet the requirements, it can be adjusted in two ways: on the one hand, adjust the initial parameter settings of the LightGBM classification model according to the cross-validation learning and prediction results and the effect of each initial parameter; on the other hand On the one hand, considering that the laboratory test and the example data of the cracking and falling karst dangerous rock project come from different environments, there may be some differences in the microseismic precursor signals. Screening, to remove samples that are incompatible with other larger samples that have multiple learning or prediction errors in the cross-validation cycle. After adjustment and re-execution of cross-validation training, the above process is repeated to finally obtain a LightGBM classification model with strong learning and generalization performance for the comprehensive identification of the stability level of the microseismic multi-precursor pull-cracking and falling karst dangerous rock.

Step 4 specific instructions:

In step 4, for the identification of the LightGBM classification model for monitoring the stability level of dangerous rock masses, steps such as data preprocessing, microseismic signal precursor feature extraction, and machine learning input feature vector construction are required. Since it is similar to step 2, LightGBM will not be described here. The input feature vector extraction process for classification models.

The present invention also provides a microseismic multi-precursor device for cracking and falling karst dangerous rock instability warning, including the following devices:

Signal acquisition unit: used for real-time acquisition of microseismic signals of cracked and fallen karst dangerous rocks;

Signal transmission unit: used to transmit the microseismic signal data of the cracked and fallen karst dangerous rock;

Signal processing unit: used for real-time preprocessing and analysis of the microseismic signals of the cracked and fallen karst dangerous rock, so as to extract various precursory characteristics of the microseismic signals of the cracked and fallen karst dangerous rock instability and collapse at each stage;

LightBGM classification model unit: used to construct the LightGBM classification model based on the cumulative apparent volume, energy fractal dimension, cumulative number of events, and b-value of the extracted microseismic signals of the karst dangerous rocks that have cracked and fallen. 4-dimensional feature vector samples, and use the sample training to establish the LightGBM classification model for the comprehensive identification of the stability level of the cracked and fallen karst dangerous rock with multiple precursors of microseismic, and perform real-time prediction of the stability level of the new cracked and fallen karst dangerous rock;

The disaster early warning unit is used to transmit the recognition results of the GPR model unit to the manager of dangerous rock mass.

Preferably, the signal processing unit includes:

Signal preprocessing subunit: used to effectively extract and denoise the microseismic information of the received cracked and fallen karst dangerous rock, so as to obtain a more concise, clean and high-quality microseismic signal;

Precursor feature extraction subunit: used to analyze the preprocessed microseismic signal in time domain, frequency domain, energy, waveform and other features to extract four types of cumulative apparent volume, energy fractal dimension, cumulative number of events and b value According to the characteristics of microseismic precursory signals, and according to the classification management rules for the microseismic precursory characteristics and stability levels of the cracked and falling karst dangerous rocks, and according to the characteristics of the four kinds of microseismic precursory characteristic indicators, they are quantified into specific danger levels.

Preferably, the LightGBM classification model unit includes:

LightGBM classification model establishment operator unit: used to construct four kinds of precursory features and stability grades based on the extracted microseismic signals, including the cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value of the karst dangerous rock that has been pulled apart and fallen. LightGBM classification model 4-dimensional feature vector sample, and use the cross-validation algorithm to train the LightGBM classification model for the comprehensive identification of the stability level of the cracking and falling karst dangerous rock with multiple precursors, and adjust the initial parameters of the LightGBM classification model according to its training and testing accuracy and training samples to obtain a LightGBM classification model with good performance;

LightGBM classification model inspection operator unit: for each sampling sample prediction result of the test sample output according to the LightGBM classification model, carry out the feasibility test of the LightGBM classification model according to the prediction result of the model test sample;

LightGBM classification model prediction operator unit: used to quantify, analyze and extract the microseismic signals of the cracked and falling karst dangerous rocks collected in real time to establish model feature vectors and input them into the LightGBM classification model. Identify the stability level of cracked and fallen karst dangerous rocks.

Compared with prior art, the beneficial effect of the present invention is:

(1) The present invention monitors the gestation, development and expansion of the cracks in the karst dangerous rock of the pull-cracking and falling type through the microseismic signal to the final failure and instability process. There are four types of microseismic signal precursor characteristic indicators, namely cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value of rock instability early warning; Complementing each other, it can better reveal the stages of the instability and collapse evolution of dangerous rock mass, and then effectively improve the early warning accuracy and early warning time of the cracking and falling karst dangerous rock instability and collapse.

(2) The present invention adopts the microseismic signal to give an early warning of the instability and collapse of the karst dangerous rock of the pulling-cracking and falling type. It differs from the objective sound and acoustic emission signals that run through the process of the instability and collapse of the dangerous rock mass in that: first, the signals of the three The collection frequency of the microseismic signal is different, and the collection frequency of the microseismic signal is less than 100Hz, which supplements the lack of low-frequency range signals in the process of the instability evolution of the dangerous rock mass in the early warning process of the latter two signals; second, the instability and collapse of the dangerous rock mass. , accompanied by the generation, development, penetration, and final overall instability and collapse of microscopic cracks, which contains rich information about large-scale microscopic cracks. In the early warning process, there is a lack of large-scale microscopic crack information in the process of instability evolution of dangerous rock mass; therefore, the method of the present invention improves the deficiencies in the collection frequency range and crack information based on microseismic and microseismic early warning methods, and further improves the cracking and falling method. Accuracy of pre-warning of instability and collapse of karst dangerous rock.

(3) The present invention comprehensively applies multiple precursory characteristic indicators to monitor and warn the whole process of cracking and falling karst dangerous rock instability and collapse. For low-level problems, through the hierarchical management of stability levels, the advanced nature of disaster warning has been significantly improved, which is conducive to prolonging the time for disaster avoidance, and thereby reducing the risk of loss of life and property caused by dangerous rock collapse disasters.

(4) The LightGBM classification machine learning model of the automatic identification of dangerous rock stability grades adopted in the present invention is a kind of reinforcement learning model based on decision tree, which has the advantages of simple and efficient implementation process and strong adaptability to high-dimensional complex problems. , and can output prediction results with probabilistic significance; it overcomes the shortcomings of the currently widely used GBDT algorithm for processing serial and high-dimensional sparse feature data that requires too much calculation cost, and can predict the karst dangerous rock precursor based on microseismic signals. The nonlinear mapping prediction problem between the characteristics and the stability grade of the cracked and fallen karst dangerous rock has strong applicability.

(5) The present invention uses a microseismic sensor to monitor the cracked and fallen karst dangerous rock in real time, obtains the microseismic signal of the dangerous rock, and transmits it to the convergence node through the sensor, and transmits it to the cloud server in real time in a unified manner through the convergence node wirelessly, and in the cloud server Carry out storage, processing and analysis, and calculate the stability level of the cracked and falling karst dangerous rock in real time. According to the calculated stability level, use the on-site alarm bell and quickly send early warning information to the user. The collapse of karst dangerous rocks caused by cracking and falling caused problems such as too slow early warning and too late to avoid them.

Description of drawings

Fig. 1 is the stress schematic diagram of the tension-cracking and falling dangerous rock provided by the content of the present invention;

Figure 2 is a schematic diagram of the accumulated apparent volume characteristics of the first microseismic signal precursor features provided by Embodiment 1 of the present invention

Fig. 3 is a schematic diagram of the energy fractal dimension characteristic of the precursor characteristic energy of the second microseismic signal provided by Embodiment 1 of the present invention

Fig. 4 is a schematic diagram of the characteristic number of events of the precursor characteristic of the third microseismic signal provided by Embodiment 1 of the present invention

Fig. 5 is a characteristic schematic diagram of the b-value characteristic of the precursor characteristic of the fourth microseismic signal provided by Embodiment 1 of the present invention

Fig. 6 is a flow chart of the LightGBM classification model establishment method for the comprehensive identification of the stability level of the microseismic multi-precursor pull-cracking and falling karst dangerous rock provided by Embodiment 1 of the present invention

Fig. 7 is a flow chart of a microseismic multi-precursor method for cracking and falling karst dangerous rock instability warning provided by Example 2 of the present invention

Figure 8 is a schematic diagram of a cloud server device provided by Embodiment 3 of the present invention

Figure 9 is a schematic diagram of a microseismic multi-precursor device for cracking and falling karst dangerous rock instability warning provided by Example 4 of the present invention

Figure 10 is a schematic diagram of a signal acquisition unit provided by Embodiment 4 of the present invention

Figure 11 is a schematic diagram of a signal transmission unit provided by Embodiment 4 of the present invention

Figure 12 is a schematic diagram of a signal processing unit provided by Embodiment 4 of the present invention

Figure 13 is a schematic diagram of a LightGBM classification model unit provided by Embodiment 4 of the present invention

Detailed ways

The specific implementation manners of the present invention will be further described below in conjunction with the accompanying drawings and examples. It should be pointed out that the drawings only show the parts related to the present invention, not all the results. And the specific examples are only to explain the present invention, not to limit the scope of the invention.

Example 1

Fig. 6 is a flow chart of a method for establishing a LightGBM classification model for the comprehensive identification of the stability grades of karst dangerous rocks with multiple precursors of microseisms and cracks and falls provided by the example of the present invention. This example can be applied to the construction of a LightGBM classification model for comprehensive identification of the stability level of karst dangerous rocks based on various precursory characteristics of microseismic signals, which include:

Step S1-1: In this embodiment 1, for the characteristics of microseismic signal precursors of cracking and falling karst dangerous rocks, four types are selected: cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value, see Table 5 for details. These four indicators contain the time domain, frequency domain, energy, waveform and other characteristics of the microseismic signal. They are independent and complementary to each other. These four indicators can better describe the whole process of the evolution process of the tearing and falling dangerous rock instability and collapse through these precursory characteristics. . Moreover, these four precursory features contain the evolution process characteristics of dangerous rock instability and collapse, that is, they can provide early warning of dangerous rock instability and collapse.

For example, see Figure 2. The cumulative apparent volume is the cumulative volume value of the rock mass in the inelastic deformation zone of the microseismic source, and the slope of its slope versus time curve is generally considered to be an important indicator of the strain rate of the rock mass, which can better It reflects the damage degree of dangerous rock mass in the process of microseismic gestation and development.

It can be calculated by the following formula:

In the formula, E is the microseismic radiant energy; μ is the shear stiffness; P is the microseismic body change potential.

According to whether the overall trend shows "initial static→slow rise→constant rise-rapid sudden increase", the whole process of rock mass instability and failure evolution can be warned. See Figure 2, a represents the initial static; b represents a slow rise; c represents a constant rise; d represents a rapid increase.

Exemplarily, referring to Fig. 3, the energy fractal dimension is a measure of the change law of the energy of the described microseismic event. When the energy change of the microseismic event is small, the fractal dimension value is in a stable low value state; otherwise, when the energy of the microseismic event changes significantly , the corresponding fractal dimension value increases rapidly. It can be calculated by the following formula:

In the formula: E is the upper limit value of the range interval of the microseismic release energy of all microseismic events; e is the microseismic release energy within the range of E; N is the logarithm value of the cumulative event number within the energy range of e; N is the microseismic event within the energy range of E total.

The whole process of rock mass instability and damage evolution can be warned according to whether the overall trend is "stable fluctuation → sudden increase point → high value stable - rapid rise"; in Figure 3, a indicates stable fluctuation; b indicates sudden increase point; c indicates a high value is stable; d indicates a rapid rise.

For example, referring to FIG. 4 , the cumulative number of events is an important parameter to measure the microseismic activity in a certain area. It can reflect the activity of microseismic signals to a certain extent, the cumulative number of events with high activity has a large change, and the cumulative number of events with low activity has a small change. It can be calculated by the following formula:

In the formula, sgn[y _i (n)] is a sign function, z(i) is the number of events in the i-th frame, and Z(i) is the cumulative number of events in the i-th frame.

According to the overall trend of the cumulative number of events, the trend of "slow rise→sudden increase point→rapid rise-accelerated rise" can reflect the evolution process of rock mass instability. In Figure 4, a represents a slow rise; b represents a sudden increase point; c represents a rapid rise; d represents an accelerated rise.

Exemplarily, referring to FIG. 5 , the b value is an important parameter used to measure the level of seismic activity in a certain region, which is the ability of the medium to control the release of accumulated energy. It can be calculated by the following formula:

In the formula, Δm is the bin interval of microseismic events, _Mi is the median number of microseismic events in bin i, and the b value is not only a function of the relative magnitude distribution of microseismic events, but also a function of the expansion scale of microseismic events.

According to whether the overall trend shows "low steady fluctuation→rapid rise→high steady fluctuation-rapid decline", the whole process of rock mass instability and failure evolution can be warned. In Figure 5, a represents; b represents a sudden increase point; c represents a stable high value; d represents a rapid rise.

Step S1-2: According to step S1-1, according to the multi-precursor characteristic indicators of the microseismic signal and the hierarchical management rules of the collapse and collapse of the karst dangerous rock instability and collapse of the cracking and falling type of karst rock, the representative laboratory tests and cracking data were collected. There are 81 examples of falling karst dangerous rocks.

Step S1-3: Firstly, the collected data of indoor test and outdoor cracked and fallen karst dangerous rock are cleaned for missing values, logical errors and unnecessary data to obtain a clean and optimized data set.

Step S1-4: According to the four kinds of microseismic signal precursor feature extraction methods in step S1-1, perform microseismic signal precursor feature extraction on the optimized data set, and quantify the stage actuality according to the obvious characteristics of each stage of the instability process of the data instance. stability level.

Step S1-5: According to the precursor characteristics of microseismic information and the actual stability level to form training samples, construct the LightGBM classification model for the comprehensive identification of the stability level of karst rocks with multiple precursors of microseismic cracking and falling.

For step S1-5, sub-steps S1-5-1, S1-5-2 and S1-5-3 are also included;

Step S1-5-1: Create training samples

In this embodiment 1, according to the multi-precursor feature data set of the microseismic signal of each sample of the cracked and fallen karst dangerous rock and its stability level, a machine learning sample (x _i , y _i ) is established, where i=1, 2,...,n, are the input feature vectors, where x _i =[x _i1 , x _i2 , x _i3 , x _i4 ,], each element is divided into cumulative apparent volume, energy fractal dimension, cumulative number of events and b value Four kinds of microseismic signals are quantitative indicators of various precursory features, and _yi is the output stability grade result of the cracked and fallen karst dangerous rock. The sample set of LightGBM classification model is as follows.

Step S1-5-2, train the LightGBM classification model

In this embodiment 1, the typical k-fold cross validation (k-fold cross validation, K-CV) method is used to randomly divide the training sample library into 10 (k=10) parts, and 9 of them are sequentially selected as training samples. The sample, and the other one as a test sample, refer to the explanation of the classification problem in the LightGBM toolbox and the existing use experience, initially set the initial parameters of the LightGBM classification model for the comprehensive identification of the stability of the microseismic multi-precursor pull-cracking and falling karst rocks, Specifically: n_estimators is 800, learning_rate is 0.15, max depth is 5, num_leaves is 32, subsample is 0.9, colsample_bytree is 0.8, min child samples is 30, njobs is 3, and according to the 10-fold cross-validation strategy, the number of training samples is N The number of all samples is 58×9/10, about 52, and the RMSE identification formula is selected as the evaluation index of the LightGBM classification model:

In the formula, y _i is the actual stability level value of the test sample, y _i* is the predicted value of the LightGBM classification model for the comprehensive identification of the stability level of the cracking and falling karst dangerous rock with multiple precursors of microseismic, N is the sample size; the test index RMSE The LightGBM classification model corresponding to the minimum value is the optimal model, and the objective formula can be expressed as min{RMSE _i }, i=1, 2, 3, . . . , m. In this embodiment, the RMSE _min is 18.73.

After several iterations, the final prediction function F* of the LightGBM classification model for the comprehensive identification of the stability level of the cracking and falling karst dangerous rock with multiple precursors of microseismic is obtained:

Exemplarily, the optimal LightGBM classification model has 2753 samples, and Table 6 only lists the information of some samples.

Preferably, in order to improve the utilization rate of a single small sample (sampling period) and optimize the training and prediction effect of the model, the present invention decomposes each waveform of a large sample of a single indoor test or engineering example to obtain a waveform of a sampling period, that is, a single The small sample is used as the sample set of the model to construct the LightGBM classification model for the comprehensive identification of the stability level of the cracking and falling karst dangerous rock with multiple precursors of microseismic, and the adjacent samples of this small sample number do not necessarily have correlation (sampling time sequence Correlation on , and correlation on sample order).

Table 6 LightGBM classification model sample set

Step S1-5-3, LightGBM classification model feasibility test

In Example 1 of the present invention, a feasibility test is performed on the results of outputting test samples from the optimal LightGBM classification model. Specifically, the test index is the prediction accuracy rate of the test sample, that is, the actual stability level of the test sample and the predicted stability level are used to check. If the prediction accuracy rate is above 95%, it is considered that the performance of the optimal LightGBM classification model established meets the requirements of Requirements, it is feasible for the prediction of the stability level of the cracked and fallen karst dangerous rock; otherwise, retrain and build the model.

Specifically, in Example 1 of the present invention, the test samples in the LightGBM classification model are finally screened by a 10-fold cross-validation algorithm (2753×1/10≈275), and the prediction accuracy of the test samples reaches 95.97%. Therefore, it is considered that the LightGBM classification model established for the comprehensive identification of the stability grades of the microseismic and multi-precursor cracked and fallen karst dangerous rocks meets the requirements, and is feasible for the early warning of the instability of the cracked and fallen karst dangerous rocks.

Exemplarily, there are 275 test samples in the LightGBM classification model, and Table 7 only lists the information of some samples. It should be noted that the number of predicted samples comes from the number in the original sample set, which is consistent with Table 7. The samples in this table There is no correlation between the numbers, which are derived from different single indoor experiments or large sample sampling periods (short time) and small samples of engineering examples.

The LightGBM classification model establishment method for the comprehensive identification of the stability grades of the cracked and fallen karst dangerous rocks with multiple precursors of microseisms provided in Example 1 is based on the hierarchical management relationship between the characteristics of the microseismic precursors and the stability grades of the selected stretched and fallen karst dangerous rocks According to the rules, extensively collect laboratory test and field example data of various precursory characteristics of microseismic signals of sampled samples at various stages of karst rock instability and collapse evolution, and use various precursory characteristic data and stability levels of microseismic data to form training samples. The typical cross-validation algorithm trains and verifies the LightGBM classification model, and obtains a LightGBM classification model with strong generalization ability and learning ability, which improves the construction efficiency of the LightGBM classification model for the comprehensive identification of the stability level of karst dangerous rocks with multiple precursors of microseismic cracking and falling And the prediction accuracy of the model, so as to improve the early warning time and accuracy of the cracking and falling karst dangerous rock instability and collapse disaster.

Table 7 LightGBM classification model test samples

Example 2

Embodiment 2 On the basis of the above-mentioned embodiment 1, a comprehensive early warning method for cracking and falling karst dangerous rock collapse based on various precursory characteristics of microseismic signals is provided. , and real-time early warning of collapse disasters. Fig. 7 is a flow chart of the comprehensive early warning method for cracking and falling karst dangerous rock collapse based on multiple precursory characteristics of microseismic signals in Embodiment 2 of the present invention. The method specifically includes the following:

Step S2-1: Example 2 of the present invention conducts real-time monitoring of the cracked and falling dangerous rocks in a mountain with a high degree of karst development in Guangxi Zhuang Autonomous Region. The rock mass is relatively complete, stable, and easy to install; then, the multiple microseismic sensors that have been arranged on the single-risk rock mass are connected to the microseismic signal collector in a network structure; finally, the microseismic signal collector is used to record And the received data collected by each microseismic sensor is transmitted to the microseismic processing system, and the present invention applies the computing module in the cloud server.

Exemplary, Embodiment 2 of the present invention, the specific installation method of the microseismic sensor: bury the microseismic monitoring sensor in the reserved hole in the rock body, bury the sensor deeply in the reserved hole with a slightly larger diameter, and use a suitable coupling The agent embeds the monitoring target and the sensor as one; the couplant has the characteristics of plasticity, quick setting, and uniformity, and removes the air and moisture in the coupling interface to realize the direct contact between the sensor and the rock wall.

Exemplarily, in Embodiment 2 of the present invention, the characteristics of the installation part of the microseismic sensor are: no excessive cracks, good contact between the interface and the microseismic sensor, high stability, and easy manual installation and disassembly.

Exemplarily, in Embodiment 2 of the present invention, in order to ensure the safety of the microseismic collector, the cables connected to the microseismic collector and the microseismic sensors are relatively long, and the installation position of the microseismic collector deviates from the cracking and falling type of karst rock failure. Stabilize the collapse-caused area.

Preferably, this embodiment considers that the main control structural plane of the karst dangerous rock mass of the split and falling type is generally in a relatively large degree of penetration, and the instability and collapse process takes a short time. In order to improve the quality of the microseismic signal generated by the internal rupture of the collected rock mass, the microseismic acquisition device selected by the present invention is a high-sensitivity piezoelectric acceleration sensor with a sampling frequency range of 0.6-300Hz and a voltage sensitivity of up to 1500mV/m /s ² .

Step S2-2: Analyze the collected microseismic data in real time in the time domain, frequency domain, energy, waveform, etc., to obtain the real-time microseismic precursor characteristic data of the cracked and fallen karst dangerous rock.

In Embodiment 2, preferably, four types are extracted: cumulative apparent volume, energy fractal dimension, cumulative event number, and b value (see Table 5 and Figures 2-5 for details).

Step S2-3: Construct a sample (x _* , y _* ) according to the various precursory characteristics of the cracking and falling karst dangerous rock microseismic microseismic characteristics and the stability level to be identified, and output it to the tearing and falling karst dangerous rock stability with multiple precursors of microseismic In the LightGBM classification model for comprehensive identification of property grades, the stability grades of real-time cracking and falling karst dangerous rocks are obtained, see Table 8 for details.

Exemplarily, there are 129 sampling samples in the whole process of cracking and falling karst dangerous rock instability and collapse warning, and Table 8 only lists the information of some samples.

Table 8 Instability early warning process of a cracked and fallen karst rock

Step S2-4: Refer to the above Table 8. It can be seen that in the process of monitoring the microseismic signal of the cracked and dropped karst dangerous rock in Example 2 of the present invention, the LightGBM classification model is used to predict the stability level of each sampling sample of the cracked and dropped karst dangerous rock , and send the early warning information to the dangerous rock manager in real time. The dangerous rock manager can identify whether the stability level identified by the LightGBM classification model reaches the Ⅳ early warning limit. The trend is consistent, such as: before identifying the stability level IV, whether the overall stability level is III; in this example, when the sample number is 128, the stability level is IV, and the previous five numbers are 123, 124, The stability grades of samples 125, 126, and 127 are all III. It is considered that the LightGBM classification model of sample No. 128 is more reliable in identifying stability grade IV. This early warning information is sent to the dangerous rock manager to determine whether to issue an early warning.

Exemplarily, considering the relatively large calculation cost and the slow expansion of cracks in the karst dangerous rock of the cracking and falling type, the sampling time of the sample sampled in Example 2 of the present invention is not fixed, and is defined according to the threshold value of its microseismic signal. If it exceeds For the threshold value, continuous sampling is performed, otherwise it is in a stagnant sampling state. This sampling method effectively reduces useless data and provides feasibility for data analysis and data wireless transmission.

Exemplarily, the sampling sample mentioned in the present invention is not a single signal point of microseismic signal collection, but refers to all the sampling data of a certain collection of microseismic signals of cracked and fallen karst dangerous rock, which is a sampling time period.

Example 3

FIG. 8 is a cloud server device proposed by the present invention, which includes one or more processors 3-1, one or more storage devices 3-2, input devices 3-3 and output devices 3-4, and these components are connected through the bus The systems 3-5 and/or other forms of linkages are interconnected. It should be noted that the components and structure of the cloud server device shown in FIG. 8 are only exemplary, not limiting, and the cloud server device may also have other components and structures as required.

The processor 3-1 can be a central processing unit (CPU) or other forms of processing units with data processing capabilities and/or instruction execution capabilities, and can control other components in the cloud server device to perform desired functions .

Exemplarily, the processor 3-1 can perform microseismic signal preprocessing, precursor feature extraction, LightGBM classification model training, prediction, and real-time early warning of cracking and falling karst dangerous rocks in the method of the present invention. The steps specifically include steps S1-3 to S1-5, and S2-2 to S2-4.

The storage device 3-2 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and/or cache memory (cache). The non-volatile memory may include, for example, a read-only memory (ROM), a hard disk, a flash memory, and the like. One or more computer program instructions can be stored on the computer-readable storage medium, and the processor 3-1 can execute the program instructions to implement the computer (implemented by the processor) in the embodiments of the present invention described below function and/or other desired functions. Various application programs and various data, such as various data used and/or generated by the application programs, may also be stored in the computer-readable storage medium.

The input device 3-3 may be a device for receiving instructions input by the user and collecting data, and its input method adopts a combination of wireless and wired transmission.

Said output device 3-4 can output various information (such as text data, image or sound) to the outside (for example user), and can comprise one or more in display, loudspeaker etc., and the application of the present invention mainly uses text data output as host.

The aforementioned input device 3-3 and output device 3-4 are mainly used for interacting with the user.

Example 4

Fig. 9 is a schematic structural diagram of a microseismic multi-precursor device for cracking and falling karst dangerous rock instability warning provided by Embodiment 4 of the present invention. This embodiment can be applied to the automatic early warning of cracking and falling karst dangerous rock disasters based on microseismic signals, and its specific structure is as follows:

Signal acquisition unit 4-1: used for real-time acquisition of microseismic signals of cracked and fallen karst rocks, and collecting the data of each sub-sensor to the control terminal; signal acquisition unit 4-1 can be included in the cloud server device shown in Figure 8 The processor 3-1 runs the program instructions stored in the storage device 3-2 to implement, and can execute step S2-1 of a microseismic multi-premonition method for cracking and falling karst dangerous rock instability warning proposed by the embodiment of the present invention;

Signal transmission unit 4-2: used to transmit microseismic signals of cracked and fallen karst dangerous rocks; signal transmission unit 4-2 can be stored by processor 3-1 in the cloud server device shown in Figure 8 and stored by storage device 3-2 It can be realized by program instructions, and can execute steps S1-2 and S2-1 of a microseismic multi-precursor method for cracking and falling karst dangerous rock instability warning proposed by the embodiment of the present invention;

Signal processing unit 4-3: used for real-time preprocessing and analysis of the microseismic signals of the cracked and fallen karst dangerous rocks, so as to extract various precursory characteristics of the microseismic signals of the cracked and fallen karst dangerous rocks at each stage of instability and collapse; the signal processing unit 4-3 can be realized by the processor 3-1 in the cloud server device shown in FIG. 8 running the program instructions stored in the storage device 3-2, and can execute a cracking and falling karst hazard proposed by the embodiment of the present invention. Steps S1-3～S1-4, S2-2 of the microseismic multi-precursor method for early warning of rock instability;

LightGBM classification model unit 4-4: used to construct 4 types of precursor characteristics and stability levels based on the extracted microseismic signals, the cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value of the cracked and fallen karst dangerous rock. LightGBM classification model 4-dimensional feature vector samples, and use the sample training to establish the LightGBM classification model for the comprehensive identification of the stability level of the cracking and falling karst dangerous rock with multiple precursors of microseismic, and carry out the real-time analysis of the stability level of the new pulling and falling karst dangerous rock identify. The LightGBM classification model unit 4-4 can be implemented by the processor 3-1 in the cloud server device shown in Figure 8 running the program instructions stored in the storage device 3-2, and can execute a split Steps S1-5 and S2-3 of the microseismic multi-precursor method for early warning of falling karst dangerous rock instability.

Disaster early warning unit 4-5: used to transmit the stability level output by the LightGBM classification model in real time to the manager of dangerous rock mass for them to judge whether to carry out early warning. The disaster early warning unit 4-5 can be realized by the processor 3-1 in the cloud server device shown in FIG. Step S2-4 of the microseismic multi-precursor method for early warning of karst dangerous rock instability.

Exemplarily, see accompanying drawing 10, described signal collection unit 4-1 comprises Fig. 10:

Signal collection sub-unit 4-1-1: used to collect the microseismic signal data of the whole process of instability and collapse evolution of each monitoring part of the cracked and fallen karst dangerous rock;

Signal acquisition control subunit 4-1-2: used to send commands to the acquisition subunits to control the microseismic signal data acquisition of each acquisition subunit. The control feature is: when the signal activity of the acquisition subunit does not exceed the set threshold value When it is in sleep mode, if it exceeds the threshold value, the acquisition mode of each acquisition sub-unit is activated, and it turns into normal mode;

For example, see accompanying drawing 11, the signal transmission unit 4-2 includes:

Signal transmission subunit 4-2-1: used to store microseismic signal data of cracked and fallen karst dangerous rocks with obvious changing characteristics, and transmit and delete them in real time;

Signal transmission control subunit 4-2-2: used to send commands to the transmission subunit to control the storage, transmission and deletion of the microseismic signal data of the transmission subunit. When the threshold value is exceeded, the storage function of the signal transmission subunit is turned on; in terms of transmission function, when the storage capacity of the microseismic signal data is greater than or equal to a complete sampling period, the transmission function of the signal transmission subunit is turned on, and the stored data is transmitted wirelessly Real-time transmission to the cloud server; delete function, when the amount of data stored in it is greater than the maximum storage capacity of the transmission sub-unit, the previous data stored in it will be gradually deleted one by one;

Exemplarily, see accompanying drawing 12, described signal processing unit 4-3 comprises:

Signal preprocessing sub-unit 4-3-1: used to effectively extract and denoise the microseismic information of the received cracked and fallen karst dangerous rock, so as to obtain relatively concise, clean and high-quality microseismic signals;

Precursor feature extraction subunit 4-3-2: used to analyze the preprocessed microseismic signal in time domain, frequency domain, energy, waveform and other features to extract cumulative apparent volume, energy fractal dimension, and cumulative event number And b value four kinds of microseismic precursor characteristics, and according to the classification management rules of the microseismic precursory characteristics and stability grades of the cracked and fallen karst dangerous rocks, and according to the characteristics of the four kinds of microseismic precursory characteristic indicators, it is quantified as a specific The danger level is divided into 1, 2, 3, 4 levels;

Exemplary, see accompanying drawing 13, described LightGBM classification model unit 4-4 comprises:

LightGBM classification model establishment operator unit 4-4-1: used for accumulative apparent volume, energy fractal dimension, accumulative number of events, and b value of 4 kinds of precursory characteristics and Stability level, construct the LightGBM classification model 4-dimensional feature vector samples, and use the cross-validation algorithm to train the LightGBM classification model for the comprehensive identification of the stability level of the microseismic multi-precursor cracking and falling karst dangerous rock, and adjust it according to its training and testing accuracy Initial parameters and training samples of the LightGBM classification model to obtain a LightGBM classification model with good performance;

LightGBM classification model inspection operator unit 4-4-2: used for performing the feasibility test of the LightGBM classification model on the model test sample prediction error according to the prediction results and prediction variance of each sampling sample of the test sample output by the LightGBM classification model;

LightGBM classification model prediction operator unit 4-4-3: used to quantify, analyze and extract the microseismic signals of the cracked and falling karst dangerous rocks collected in real time to establish the model feature vector and input it to In the LightGBM classification model, identify the stability level of the cracked and fallen karst dangerous rock;

Exemplarily, the devices mentioned in the present invention can be implemented according to the processor 3-1 in the cloud server device in FIG. 8 running the program instructions stored in the storage device 3-2. A subunit does not mean that the processing of this device is not involved.

Exemplarily, since the device proposed by the present invention is an intelligent control type microseismic signal acquisition device, it does not collect real-time acquisition with high power consumption for a long time around the clock, but has an intelligent acquisition device with a trigger mechanism, which can be understood as: the signal is low and active When the signal is high, the acquisition device is in the open state, and the signal is collected in real time; therefore, the abscissa of the accompanying drawings 2 to 5 of the present invention-sampling sample does not represent a continuous sampling time , but represents an intermittent sampling time, and a sampling sample represents a sampling of a valid signal.

Those of ordinary skill in the art can understand that all or part of the steps carried by the methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. During execution, one or a combination of the steps of the method embodiments is included.

In addition, each subunit in each embodiment of the present application may be integrated into one unit, each subunit may exist separately physically, or two or more subunits may be integrated into one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units. If the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

It should be noted that the purpose of announcing the above examples is to help further understand the present invention, but those skilled in the art can understand that if various obvious changes, readjustments and alternative means are made to the present invention, they will not depart from the protection scope of the present invention . Therefore, the present invention is not limited to the content disclosed in the examples, and the protection scope of the present invention is subject to the scope defined in the claims.

Claims (5)

1. The microseismic multi-premonition method for the early warning of the instability of the karst dangerous rock that is pulled and dropped is characterized in that it includes the following steps: Step 1: In order to effectively monitor the instability and collapse process of the karst dangerous rock that is pulled and dropped, the microseismic signal with a frequency of 0-100 Hz generated by the large-scale rupture inside the dangerous rock is used as the monitoring information, and the accumulated apparent volume, energy fractal dimension, The cumulative number of events and the b value of four kinds of microseismic precursor characteristic indicators are used as early warning indicators of dangerous rock instability, and the relationship rules between the four kinds of microseismic precursory characteristic indicators and the instability of karst rocks of the split and fall type are formulated respectively, and then the split and fall formula is formulated. The rules of the comprehensive relationship between the four types of microseismic precursory characteristics of karst dangerous rocks and the stability grades of dangerous rocks; Step 2: From laboratory tests and on-site engineering microseismic monitoring examples, collect the characteristics of four types of microseismic signal precursors in each stage of the collapse and collapse of karst dangerous rock instability and collapse, and establish a machine learning sample set; Step 3: Using the machine learning sample set, according to the cross-validation strategy, train the LightGBM classification machine learning model that has excellent adaptability to processing parallel and large-scale high-dimensional data problems, and use the trained LightGBM classification model to construct four microseismic precursor features The nonlinear mapping relationship between indicators and the stability level of dangerous rocks; Step 4: For the cracked karst dangerous rock mass that needs to be monitored and warned, the data collected by the microseismic equipment is used to obtain the corresponding scores of the four types of microseismic precursor features according to the scoring rule table, and a 4-dimensional feature index vector is constructed. It is input to the trained LightGBM classification model to obtain the model prediction result of the cracked and fallen karst dangerous rock mass; Wherein, the feature of extracting the energy fractal dimension is: calculating the correlation integral C(e) of the energy distribution of the microseismic event produced by the microfracture of the rock mass; by taking lge as the abscissa and lgC(e) as the ordinate, a rectangular coordinate system is established , the characteristic of the energy fractal dimension De is obtained by fitting, and e represents the microseismic release energy; Wherein, the feature of extracting the accumulated event number is: setting threshold value threshold value of microseismic signal waveform amplitude according to various factors such as monitoring object, environmental condition and load condition; The time-invariant microseismic signals are filtered by the threshold value to obtain the number of events; finally, they are accumulated and summed to obtain the required cumulative event number characteristics; Wherein, the b-value extraction feature is: setting the microseismic signal optimal event bin interval Δm; calculating the median number M _i of each microseismic event; using the linear least square method to calculate the b-value feature; its calculation formula is:

Among them, N is the total number of microseismic events, and i is the serial number of microseismic events. 2. The microseismic multi-premonition method of pulling and falling type karst dangerous rock instability pre-warning according to claim 1, is characterized in that, the cumulative apparent volume extraction feature of the microseismic signal precursory feature is: first calculate rock mass cracking and friction slip The microseismic radiation energy E caused by the transformation of the rock mass from elastic deformation to inelastic deformation during the migration process; then calculate the volume change of the rock mass in the inelastic deformation area at the source; the microseismic body potential P; finally use the microseismic radiation energy E The variable potential P solves the volume V _A , and performs accumulative superposition processing on it to obtain the feature of the accumulative apparent volume _ΣVA . 3. The microseismic multi-precursor device of the microseismic multi-premonitory device for the cracking and falling type karst dangerous rock instability warning based on the microseismic multi-precursor method of claim 1, it is characterized in that, comprising: Signal acquisition unit: used for real-time acquisition of microseismic signals of cracked and fallen karst dangerous rocks; Signal input unit: used to transmit microseismic signal data of cracked and fallen karst dangerous rocks; Signal processing unit: used for real-time preprocessing and analysis of the microseismic signals of the cracked and fallen karst dangerous rock, so as to extract various precursory characteristics of the microseismic signals of the cracked and fallen karst dangerous rock instability and collapse at each stage; LightGBM classification model unit: used to construct the LightGBM classification model based on the cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value of the four precursory characteristics and stability levels of the extracted microseismic signal of the karst rock that has cracked and fallen 4-dimensional feature vector samples, and use the sample training to establish a LightGBM classification model for the comprehensive identification of the stability level of the cracking and falling karst dangerous rock with multiple precursors of microseismic, and carry out real-time identification of the stability level of the new cracking and falling karst dangerous rock; Disaster early warning unit: used to transmit the recognition result of the LightGBM classification model to the dangerous rock manager. 4. The microseismic multi-precursor device for cracking and falling karst dangerous rock instability warning according to claim 3, characterized in that, the signal processing unit includes: Signal preprocessing subunit: used to effectively extract and denoise the microseismic information of the received cracked and fallen karst dangerous rock, so as to obtain a more concise, clean and high-quality microseismic signal; Precursor feature extraction subunit: used to analyze the preprocessed microseismic signal in time domain, frequency domain, energy, and waveform characteristics to extract four types of microseismic cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value According to the classification management rules for the microseismic precursory characteristics and stability levels of the cracked and falling karst dangerous rocks, and according to the characteristics of the four types of microseismic precursory characteristic indicators, they are quantified into specific hazard levels. 5. The microseismic multi-precursor device for cracking and falling karst dangerous rock instability warning according to claim 3, wherein the LightGBM classification model unit includes: LightGBM classification model establishment operator unit: used to construct four kinds of precursory features and stability grades based on the extracted microseismic signals, including the cumulative apparent volume, energy fractal dimension, cumulative number of events, and b value of the karst dangerous rock that has been pulled apart and fallen. LightGBM classification model 4-dimensional feature vector sample, and use cross-validation algorithm to train LightGBM classification model for comprehensive identification of microseismic multi-precursor cracking and falling karst rock stability grades, and adjust the initial parameters of LightGBM classification model according to its training and testing accuracy. Training samples to obtain a LightGBM classification model with good performance; LightGBM classification model inspection operator unit: used to predict the results of each sampling sample according to the test samples output by the LightGBM classification model, and carry out the feasibility test of the LightGBM classification model according to the prediction results of the model test samples; LightGBM classification model prediction operator unit: used to quantify, analyze and extract the microseismic signals of the cracked and falling karst dangerous rocks collected in real time to establish model feature vectors and input them into the LightGBM classification model. Identify the stability level of cracked and fallen karst dangerous rocks.

Images