Blasting vibration monitoring and influence range prediction method for long and large tunnel in environment sensitive area

Abstract

The invention provides a blasting vibration monitoring and influence range prediction method for a long and large tunnel in an environment sensitive area, and relates to the technical field of blasting safety monitoring and vibration control in tunnel engineering, and the method comprises the steps: constructing a three-dimensional dynamic monitoring network, collecting multi-source data based on the three-dimensional dynamic monitoring network, carrying out the preprocessing of the multi-source data, obtaining a blasting vibration feature matrix, and carrying out the prediction of the influence range. Building a prediction model based on the improved LSTM model, training the prediction model based on a preset blasting vibration characteristic matrix and blasting parameters and geological conditions corresponding to the preset blasting vibration characteristic matrix to obtain a trained prediction model, and inputting a to-be-detected blasting vibration characteristic matrix and blasting parameters and geological conditions corresponding to the to-be-detected blasting vibration characteristic matrix into the trained prediction model. And obtaining a prediction result. According to the method, the precision and the real-time performance of blasting vibration monitoring and prediction are remarkably improved, the vibration control problem in blasting construction of the long and large tunnel in the environment sensitive area is effectively solved, the prediction precision is high, the real-time performance is high, the adaptability is high, and the data utilization efficiency is high.

Description

Technical Field

[0001] The present invention relates to the technical field of blasting safety monitoring and vibration control in tunnel engineering, and particularly to a method for monitoring blasting vibration and predicting the influence range of a long tunnel in an environmentally sensitive area. Background Art

[0002] Tunnel blasting construction is one of the main means of tunnel excavation, but the vibration generated by it will have an adverse impact on the surrounding environment (such as buildings, cultural relics, residential areas, etc.). At present, traditional empirical formula methods (Sadovsky formula, Hopkinson bar model) are mainly used for tunnel blasting vibration monitoring, and there are the following deficiencies:

[0003] 1. Low prediction accuracy: The traditional method does not consider the dynamic coupling effect of geological conditions and blasting parameters, resulting in a large prediction error; it cannot meet the requirements of millimeter-level vibration control in environmentally sensitive areas.

[0004] 2. Poor real-time performance: Existing methods are difficult to achieve real-time prediction of the vibration influence range; there is a lack of a real-time grading early warning mechanism for sensitive targets.

[0005] 3. Weak adaptability: Traditional models rely on historical data and have poor adaptability to new geological sections; they cannot dynamically adjust model parameters to adapt to complex geological conditions.

[0006] 4. Insufficient data utilization: Existing methods do not fully utilize multi-source data (such as vibration, acoustic emission, stress data); there is a lack of in-depth mining and fusion analysis of monitoring data.

[0007] Therefore, it is very necessary to design a method for monitoring blasting vibration and predicting the influence range of a long tunnel in an environmentally sensitive area. Summary of the Invention

[0008] In order to overcome the deficiencies of the prior art, the purpose of the present invention is to provide a method for monitoring blasting vibration and predicting the influence range of a long tunnel in an environmentally sensitive area.

[0009] To achieve the above purpose, the present invention provides the following solutions:

[0010] The present invention provides a method for monitoring blasting vibration and predicting the influence range of a long tunnel in an environmentally sensitive area, including:

[0011] Step 1: Construct a three-dimensional dynamic monitoring network;

[0012] Step 2: Collect multi-source data based on the three-dimensional dynamic monitoring network and preprocess it to obtain a blasting vibration feature matrix;

[0013] Step 3: Build a prediction model based on the improved LSTM model, and train the prediction model based on the preset blasting vibration feature matrix and its corresponding blasting parameters and geological conditions to obtain the trained prediction model;

[0014] Step 4: Input the blasting vibration feature matrix to be detected and its corresponding blasting parameters and geological conditions into the trained prediction model to obtain a prediction result.

[0015] Preferably, in Step 1, a three-dimensional dynamic monitoring network is constructed, specifically:

[0016] Based on the UAV cluster, monitoring terminals are arranged at regular intervals along the tunnel axis, and dual-redundant monitoring terminals are set around sensitive targets. Real-time data transmission is realized based on LoRa and 5G dual-mode communication. Among them, the monitoring terminals are equipped with vibration sensors, acoustic emission sensors and stress sensors.

[0017] Preferably, in Step 2, multi-source data is collected based on the three-dimensional dynamic monitoring network and preprocessed to obtain a blasting vibration feature matrix, specifically:

[0018] Step 201: Collect the original vibration signal based on the vibration sensor, collect the original acoustic emission signal based on the acoustic emission sensor, and collect the original stress signal based on the stress sensor;

[0019] Step 202: Perform signal denoising processing on the collected original vibration signal, original acoustic emission signal and original stress signal;

[0020] Step 203: Perform data normalization processing and time alignment processing on the vibration signal, acoustic emission signal and stress signal after signal denoising processing;

[0021] Step 204: Extract the features of the processed vibration signal, acoustic emission signal and stress signal, and construct a blasting vibration feature matrix.

[0022] Preferably, in Step 202, signal denoising processing is performed on the collected original vibration signal, original acoustic emission signal and original stress signal, specifically:

[0023] Based on the improved wavelet threshold denoising algorithm, perform denoising processing on the original vibration signal, original acoustic emission signal and original stress signal. The specific process is as follows:

[0024] Discretize the original vibration signal, original acoustic emission signal and original stress signal, perform 4-layer decomposition on the wavelet, and obtain the approximate component and detail component of each layer of decomposition and the amplitude of the wavelet detail signal of this layer;

[0025] Calculate the noise standard deviation and adaptive threshold of each layer of signal according to the detail signal coefficients of each layer of signal;

[0026] Process the signal coefficients of each layer using a threshold quantization function to complete threshold noise reduction at each scale;

[0027] Perform an inverse wavelet transform on the processed components of each layer to obtain the noise-reduced components and complete wavelet reconstruction.

[0028] Preferably, in step 204, extract the characteristics of the processed vibration signal, acoustic emission signal, and stress signal, and construct a blasting vibration characteristic matrix, specifically:

[0029] For the vibration signal, extract its vibration characteristics, including the peak particle vibration velocity, dominant vibration frequency, and vibration duration;

[0030] For the acoustic emission signal, extract its acoustic emission characteristics, including the acoustic emission event count, acoustic emission energy, and acoustic emission amplitude;

[0031] For the stress signal, extract its stress characteristics, including the peak stress, stress change rate, and stress recovery time;

[0032] Based on the vibration characteristics, acoustic emission characteristics, and stress characteristics, construct a blasting vibration characteristic matrix F, which is:

[0033] F = [PPV, dominant vibration frequency, duration, acoustic emission event count, acoustic emission energy, acoustic emission amplitude, peak stress, stress change rate, stress recovery time].

[0034] Preferably, in step 3, build a prediction model based on an improved LSTM model, and train the prediction model based on a preset blasting vibration characteristic matrix and its corresponding blasting parameters and geological conditions to obtain the trained prediction model, specifically:

[0035] Build a prediction model based on the LSTM-SA model, where the LSTM-SA model consists of a Self-Attention calculation module, an LSTM module, and a fully connected neural network;

[0036] During its training process, use the SGDR learning rate adjustment algorithm for training.

[0037] Preferably, the blasting parameters include the charge amount, burden, blasting method, and delay time, and the geological conditions include the rock mass grade, fracture density, rock mass wave velocity, and groundwater condition.

[0038] According to the specific embodiments provided by the present invention, the following technical effects are disclosed by the present invention:

[0039] The present invention provides a method for monitoring blasting vibration and predicting the influence range of long tunnels in environmentally sensitive areas. The method includes: constructing a three-dimensional dynamic monitoring network, collecting multi-source data based on the three-dimensional dynamic monitoring network, and preprocessing it to obtain a blasting vibration feature matrix; building a prediction model based on an improved LSTM model; training the prediction model based on a preset blasting vibration feature matrix and its corresponding blasting parameters and geological conditions to obtain a trained prediction model; and inputting the blasting vibration feature matrix to be detected and its corresponding blasting parameters and geological conditions into the trained prediction model to obtain a prediction result. The present invention has the following advantages:

[0040] 1. The prediction accuracy is improved. By using multi-source data fusion, combining multi-modal data such as vibration, acoustic emission, and stress, a blasting vibration feature matrix is constructed. By improving the input quality of the prediction model through data fusion, the model can be dynamically corrected, significantly reducing errors, and still maintaining a high prediction accuracy under complex geological conditions;

[0041] 2. Real-time monitoring and early warning are realized. Real-time data collection can be carried out, and real-time transmission of monitoring data is achieved by using LoRa+5G dual-mode communication. The prediction result of the vibration influence range can be updated every 5 seconds. A hierarchical early warning mechanism is adopted: a three-level early warning system is established to trigger early warning signals in a timely manner;

[0042] 3. The data utilization efficiency is improved. By using multi-modal data fusion, integrating multi-source data such as vibration, acoustic emission, and stress, a blasting vibration feature matrix is constructed. By extracting key features and deeply mining the data value, and through noise reduction processing, the data quality is improved;

[0043] 4. Environmentally sensitive targets can be protected. The vibration influence range can be accurately predicted, avoiding damage to sensitive targets, being easy to implement, reducing construction costs, and reducing unnecessary vibration reduction measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0045] Figure 1 It is a flowchart of the method provided by the embodiment of the present invention;

[0046] Figure 2 It is a schematic diagram of the improved wavelet threshold denoising process;

[0047] Figure 3 It is a schematic diagram of the LSTM-SA model structure. Detailed implementation manners

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0049] The object of the present invention is to provide a method for monitoring blasting vibration and predicting the influence range of long tunnels in environmentally sensitive areas, which significantly improves the accuracy and real-time performance of blasting vibration monitoring and prediction, effectively solves the vibration control problem in the blasting construction of long tunnels in environmentally sensitive areas, and has high prediction accuracy, strong real-time performance, strong adaptability, and high data utilization efficiency.

[0050] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners.

[0051] Figure 1 For the method flow chart provided by the embodiment of the present invention, as Figure 1 shown, the present invention provides a method for monitoring blasting vibration and predicting the influence range of long tunnels in environmentally sensitive areas, including:

[0052] Step 1: Construct a three-dimensional dynamic monitoring network;

[0053] Step 2: Collect multi-source data based on the three-dimensional dynamic monitoring network, and preprocess it to obtain a blasting vibration feature matrix;

[0054] Step 3: Build a prediction model based on an improved LSTM model, and train the prediction model based on a preset blasting vibration feature matrix and its corresponding blasting parameters and geological conditions to obtain a trained prediction model;

[0055] Step 4: Input the blasting vibration feature matrix to be detected and its corresponding blasting parameters and geological conditions into the trained prediction model to obtain a prediction result.

[0056] In Step 1, constructing a three-dimensional dynamic monitoring network specifically includes:

[0057] Arranging monitoring terminals at regular intervals along the tunnel axis based on an unmanned aerial vehicle (UAV) cluster, setting dual-redundant monitoring terminals around sensitive targets, and realizing real-time data transmission based on LoRa and 5G dual-mode communication. Among them, the monitoring terminals are provided with vibration sensors, acoustic emission sensors, and stress sensors.

[0058] In Step 2, collecting multi-source data based on the three-dimensional dynamic monitoring network and preprocessing it to obtain a blasting vibration feature matrix specifically includes:

[0059] Step 201: Collect the original vibration signal based on a vibration sensor, collect the original acoustic emission signal based on an acoustic emission sensor, and collect the original stress signal based on a stress sensor;

[0060] Step 202: Perform signal denoising processing on the collected original vibration signal, original acoustic emission signal, and original stress signal;

[0061] Step 203: Perform data normalization processing and time alignment processing on the vibration signal, acoustic emission signal, and stress signal after signal denoising processing;

[0062] Step 204: Extract the features of the processed vibration signal, acoustic emission signal, and stress signal, and construct a blasting vibration feature matrix.

[0063] In Step 202, when performing signal denoising processing on the collected original vibration signal, original acoustic emission signal, and original stress signal, specifically:

[0064] Perform denoising processing on the original vibration signal, original acoustic emission signal, and original stress signal based on an improved wavelet threshold denoising algorithm. First, introduce the improved wavelet threshold denoising algorithm:

[0065] The wavelet threshold is a time-scale analysis denoising method for signals, with the characteristics of multi-resolution analysis, and has the ability to represent the local characteristics of signals in both the time-frequency domains. It is a time-frequency localization analysis method with a fixed window size but can change its shape, and both the time window and frequency window can be changed. That is, it has a lower time resolution and a higher frequency resolution in the low-frequency part, and a higher time resolution and a lower frequency resolution in the high-frequency part, which is very suitable for analyzing non-stationary signals and extracting the local characteristics of signals. The wavelet coefficients generated by the signal contain important information. After wavelet decomposition, the wavelet coefficients of the signal are larger, and the wavelet coefficients of the noise are small. The wavelet coefficients of the noise should be smaller than those of the signal. By selecting an appropriate threshold, the wavelet coefficients greater than the threshold are considered useful signals and should be retained, while those smaller than the threshold are noise and are set to zero. Finally, reconstruction is performed to achieve the purpose of denoising;

[0066] Introduce its principle:

[0067] Generally, a one-dimensional noisy signal can be expressed as f(t) = s(t) + n(t), where s(t) is the useful signal and n(t) follows N(0,σ 2) The noise signal, in most cases, the signal is mainly concentrated in the low-frequency part, and the noise is distributed in the high-frequency part. Therefore, after wavelet decomposition, the wavelet transform coefficients of the microseismic signal are greater than those of the noise. During the processing, a suitable threshold is generally selected. When the wavelet coefficient is less than this value, it indicates that the wavelet is generated by noise. On the contrary, it is mainly caused by the signal. Threshold denoising is performed on the sub-band coefficients with noise, and the wavelet inverse transform is performed on the high-frequency components after denoising and the low-frequency components where the signal is concentrated to complete the reconstruction, that is, the noise interference unrelated to the microseismic signal can be effectively removed. So the basic steps of wavelet threshold denoising are as follows:

[0068] 1. Decomposition: Select the wavelet and determine the number of decomposition levels of the wavelet.

[0069] 2. Threshold processing stage: Select the threshold and the threshold function to perform non-linear processing on the wavelet coefficients.

[0070] 3. Reconstruction: Use the new wavelet coefficients to reconstruct the signal to obtain the filtered signal.

[0071] The selection of the wavelet threshold is specifically as follows:

[0072] Four commonly used thresholds in wavelet threshold analysis include the fixed-form threshold, the adaptive threshold, the heuristic threshold, and the minimax threshold.

[0073] Fixed-form threshold. The threshold of this method is a fixed value, and its value formula is:

[0074]

[0075] In the formula, σ is the standard deviation of the noise signal, and N is the signal length.

[0076] The threshold is always fixed, but the modulus maximum value of the noise signal decreases as the decomposition level increases. If the same threshold is applied to different decomposition levels, too much useful signal will be filtered out in the low-frequency coefficients, and part of the noise will be retained in the high-frequency coefficients.

[0077] Adaptive threshold, based on Stein's unbiased risk estimate:

[0078] Take the absolute value of each element in the signal s(i), arrange them in order, and then square each element to obtain a new signal sequence:

[0079] f(k) = [sort(|s|)] 2 , (k = 0, 1,..., N - 1)

[0080] If the threshold is taken as the square root of the k-th element of f(k), that is

[0081]

[0082] The risk generated by this threshold is

[0083]

[0084] According to the obtained risk curve Rish(k), denote the value corresponding to its minimum risk point as k min , then the adaptive threshold is defined as:

[0085]

[0086] During the value-taking process, the computational amount increases, the operation time is long, and the noise reduction effect is average;

[0087] Heuristic threshold. This method is a compromise between the fixed threshold and the adaptive threshold, and its value-taking formula is:

[0088]

[0089] In the formula: N is the signal length, and S is the number of coefficients;

[0090] 4. Minimax threshold, and its value-taking formula is:

[0091]

[0092] In the formula, σ = median(|x|) / 0.6745;

[0093] This method also belongs to the fixed threshold method. It can minimize the mean square error under the worst conditions, but the value-taking is extreme and it is easy to cause the loss of useful signal information;

[0094] The selection of the threshold is another important step in wavelet threshold denoising. If the selected threshold is too large, some of the original signals will be misidentified as noise and some useful signals will be filtered out. If the threshold is too small, there will be too much noise signal in the wavelet coefficients, resulting in signal distortion and a decline in the noise reduction effect. Therefore, the threshold selection will also affect the quality of filtering;

[0095] Therefore, in the denoising algorithm proposed by the present invention, after the threshold selection experiment, the following improved threshold determination method is adopted:

[0096]

[0097] In the formula, j is the decomposition scale. It can be seen that since the threshold of the noise wavelet coefficients decreases as the decomposition scale increases, noise and signals can be more effectively distinguished, and signals can be retained to a greater extent while removing noise;

[0098] The selection of the wavelet threshold function is specifically:

[0099] Threshold denoising is to obtain the high-frequency coefficients of each layer and the lowest low-frequency coefficient by wavelet decomposition of the signal. Generally speaking, the high-frequency coefficient is the original signal, while the low-frequency coefficient is the noise. A threshold function is needed to filter the low-frequency coefficient with noise and remove the noise coefficient. The traditional hard and soft threshold functions proposed by Donoho and the common threshold selection functions are as follows:

[0100]

[0101] The y estimated by the traditional soft threshold method s (x) is a continuous function, but there is a fixed constant between the absolute value and the true value (when |x|≥λ). This difference will cause an error in the value. It may not be the best to completely remove this deviation during the hard threshold processing. Correspondingly, the function is continuous but not differentiable at ±λ, and there is a step phenomenon. The corresponding part of the noise component in x still needs to be removed;

[0102] Considering the above problems, an improved threshold function is designed. The threshold function is not only continuous with the soft threshold function, but also when |x| increases, the deviation between y(x) and |x| gradually decreases. The new threshold function proposed by the present invention is as follows:

[0103]

[0104] In the formula, j is the decomposition scale;

[0105] When |x|→λ, y p (x)→0, indicating that this function is similar to the soft threshold function and is continuous like the soft threshold function without a step phenomenon. When |x|→∞, y p (x)→x. At this time, the new threshold function is similar to the hard threshold function, solving the problem that there is always a certain deviation in the design of the soft threshold function. In addition, it can be found that the improved threshold function changes with the change of the decomposition scale j, enhancing the adaptability to different decomposition layers. There are no uncertain parameters in the function, so the stability of signal denoising is improved;

[0106] In summary, as Figure 2 shown, the specific denoising process of the present invention is as follows:

[0107] Input: the original vibration signal, the original acoustic emission signal and the original stress signal x(t), t = 1, 2,..., N;

[0108] Output: the denoised signal y(t), t = 1, 2,..., N;

[0109] 1. Discretize the original vibration signal, original acoustic emission signal, and original stress signal x(t), and perform 4-layer decomposition on the wavelet to obtain the approximate component {CA j,k} and the detail component {CD j,k} of each layer of decomposition, as well as the amplitude A j of the wavelet detail signal of this layer;

[0110]

[0111] 2. Calculate the noise standard deviation σ j and the adaptive threshold thr j of each layer according to the detail signal coefficients of the signals of each layer;

[0112]

[0113] 3. Use the threshold quantization function to process the signal coefficients of each layer to complete the threshold denoising at each scale;

[0114] 4. Perform wavelet inverse transform on the processed components of each layer to obtain the denoised components and complete the wavelet reconstruction;

[0115]

[0116] The algorithm process ends.

[0117] In step 204, extract the characteristics of the processed vibration signal, acoustic emission signal, and stress signal, and construct a blasting vibration characteristic matrix, specifically:

[0118] For the vibration signal, extract its vibration characteristics, including:

[0119] Peak particle vibration velocity (PPV): Record the maximum vibration velocity caused by blasting (unit: mm / s);

[0120] Dominant vibration frequency: Extract the main frequency of the vibration signal through Fourier transform (unit: Hz);

[0121] Vibration duration: The time from the start of the vibration signal to the attenuation to 10% of the peak value (unit: ms);

[0122] For the acoustic emission signal, extract its acoustic emission characteristics, including:

[0123] Acoustic emission event count: Record the total number of acoustic emission events generated by the rock mass rupture during blasting;

[0124] Acoustic emission energy: Calculate the total energy of the acoustic emission signal (unit: aJ);

[0125] Acoustic emission amplitude: Extract the maximum amplitude of the acoustic emission signal (unit: dB);

[0126] For the stress signal, extract its stress characteristics, including:

[0127] Peak stress: Record the maximum stress value caused by blasting (unit: MPa);

[0128] Stress change rate: Calculate the change rate of stress over time (unit: MPa / s);

[0129] Stress recovery time: The time for the stress to recover from the peak to the stable state (unit: s);

[0130] Based on the vibration characteristics, acoustic emission characteristics and stress characteristics, construct the blasting vibration characteristic matrix F, which is:

[0131] F = [PPV, main vibration frequency, duration, acoustic emission event count, acoustic emission energy, acoustic emission amplitude, peak stress, stress change rate, stress recovery time].

[0132] In step 3, build a prediction model based on the improved LSTM model, and train the prediction model based on the preset blasting vibration characteristic matrix and its corresponding blasting parameters and geological conditions to obtain the trained prediction model. Specifically:

[0133] Build a prediction model based on the LSTM-SA model, where the LSTM-SA model consists of a Self-Attention calculation module, an LSTM module and a fully connected neural network;

[0134] The LSTM-SA model consists of a self-attention (Self-Attention) calculation module, an LSTM module and a fully connected neural network. The model structure is as Figure 3 shown. The LSTM-SA model first calculates the self-attention weight vector of the data at each moment in the input sequence by applying the self-attention mechanism. In this step, the data at each moment is calculated with the data at all other moments to obtain the correlation between the two, and a series of weight coefficients are generated to form the attention weight vector. In order to combine the LSTM with the self-attention mechanism to alleviate the memory decay problem existing in the LSTM, the LSTM-SA model concatenates the original input at the current moment with the attention weight vector at the corresponding moment and uses it as the input of the LSTM module;

[0135] By connecting the self-attention weight with the original input, the LSTM-SA model can utilize the advantages of the self-attention mechanism to dynamically adjust the importance of the input data. This way can help the model better capture the correlation and dependency relationships at different moments in the input sequence, thereby improving the model's expression ability and prediction performance;

[0136] In this model, the self-attention weight calculation module uses a multi-layer perceptron method to calculate the correlation weights. The LSTM module consists of two stacked LSTM layers, with 64 and 32 neurons respectively. The fully connected network consists of two fully connected layers, with 32 and 1 neurons respectively. Its main function is to integrate the output of the LSTM module and map it to a health indicator;

[0137] In the selection of activation functions, the ReLu function is used as the activation function in both the self-attention weight calculation module and the LSTM module. In the last layer of the fully connected network, the Sigmoid function is used to limit the output value within the range of [0, 1];

[0138] During its training process, the SGDR learning rate adjustment algorithm is used for training, specifically as follows:

[0139] During the neural network training process, many hyperparameters need to be adjusted to make the model achieve the best convergence effect. The learning rate is one of the most significant hyperparameters affecting network training. In neural network training, the learning rate represents the step size for the model to move towards the direction of the minimum error. In each iteration, the neural network calculates the predicted value through forward calculation based on the input data. After calculating the loss error with the label value, the error is backpropagated to adjust the network weights, and the iteration is repeated to gradually make the network model reach the global optimal solution. The learning rate is the parameter of the weight adjustment step size in each iteration, usually denoted by n. The update formula for the network weights is:

[0140]

[0141] An appropriate learning rate can help the network quickly converge to the optimal solution. However, if the learning rate is set too large or too small, the opposite result will be obtained. If the learning rate is too large, the gradient update will oscillate near the optimal solution but cannot reach the optimal solution. From the result, it means that the network is difficult to converge, and the prediction accuracy fluctuates. If the learning rate is too small, although the model updates the gradient in the direction of the optimal solution, the update speed is slow. From the result, it means that the network converges slowly. On the other hand, if the learning rate is too small, it may also cause the model to fall into a local optimal solution, that is, a saddle point, and cannot reach the global optimal solution;

[0142] Therefore, the setting of the learning rate determines the difficulty of training the network model and directly affects the model accuracy. In the present invention, an algorithm for dynamically adjusting the learning rate is used. Common dynamic learning rate adjustment strategies include:

[0143] Step decay: After the network iteratively trains to a specified number of times, multiply the learning rate by a preset coefficient;

[0144] Exponential decay: During the network iterative training process, the learning rate decreases in the form of an exponential function;

[0145] Cyclic learning rate: A method of cyclically adjusting the learning rate, which allows the learning rate to gradually increase from a minimum value to a maximum value and repeat this process, causing the learning rate to change periodically.

[0146] Cosine annealing learning rate: During the training process, the learning rate decreases from the maximum value to the minimum value in the form of a cosine function. The rate of decrease is slow at first, then accelerates, and finally slows down again. Such a decreasing pattern can bring good results to network training.

[0147] The objective optimization function of a neural network is a convex function. Therefore, during training, when the model searches for the optimal solution in the direction of the minimum error, it may fall into a local optimal solution. By increasing the learning rate to make the model jump out of the local optimal solution, this method is called Stochastic Gradient Descent with Restarts (SGDR).

[0148] The present invention applies the SGDR learning rate adjustment algorithm to the training of the LSTM-SA model. Its advantage is that it can force the model to jump out of the local optimal point by suddenly increasing the learning rate at each cycle of the cyclic change of the learning rate. However, at the beginning of each cycle, the learning rate will suddenly increase to the preset maximum value, which will cause oscillations in the model during the training process and is not conducive to the stable convergence of the model. Therefore, the present invention changes the sudden increase of the initial learning rate to the maximum value at each cycle to a linear increase to the maximum value, reducing the problem of model oscillation.

[0149] From the above analysis, it can be seen that in the Stochastic Gradient Descent with Restarts (SGDR) algorithm, the learning rate changes periodically within the range of the maximum and minimum values, but the algorithm does not give a method to determine the boundary values. Therefore, the present invention first conducts multiple pre-trainings on the training set, searches for the optimal boundary values through pre-training on the training data set. During the pre-training process, the initial value of the learning rate is set to a very small value, and the learning rate is increased exponentially to a larger value at each iteration. During this process, the values of the learning rate and the loss error are recorded. The learning rate at which the loss error starts to decrease is used as the lower limit of SGDR, and the learning rate at which the loss error starts to increase is used as the upper limit of SGDR. This method can keep the periodic change of the learning rate within the interval that makes the loss error smaller, making the training process more stable and also accelerating the convergence speed of the model.

[0150] The input of the prediction model is the blasting parameters, geological conditions, and blasting vibration characteristic matrix.

[0151] The output of the prediction model is:

[0152] Vibration influence range: The range of the area where the predicted vibration velocity exceeds the threshold (unit: m).

[0153] Vibration intensity distribution: The predicted vibration velocity (PPV) values at different positions (unit: mm / s).

[0154] The blasting parameters include:

[0155] Charge amount (Q): The amount of explosive for single - stage blasting (unit: kg);

[0156] Burden (W): The minimum distance from the charge center to the free face (unit: m);

[0157] Blasting method: Such as hole - by - hole initiation, row - by - row initiation, etc.;

[0158] Delay time: The inter - stage delay time (unit: ms);

[0159] The geological conditions include:

[0160] Rock mass grade (R): Divided according to RQD (Rock Quality Designation) or rock mass integrity coefficient;

[0161] Fracture density (D): The number of fractures per unit volume (unit: fractures / m 3 );

[0162] Rock mass wave velocity (Vp): The propagation velocity of longitudinal waves in the rock mass (unit: m / s);

[0163] Groundwater condition: Such as dry, moist or saturated.

[0164] In step 4, input the blasting vibration characteristic matrix to be detected, together with its corresponding blasting parameters and geological conditions, into the trained prediction model to obtain a prediction result. Specifically:

[0165] Input the blasting vibration characteristic matrix to be detected, together with its corresponding blasting parameters and geological conditions, into the trained prediction model to obtain a prediction result. According to the prediction result, conduct risk level classification. For example:

[0166] Low risk: PPV < 50% of the limit value;

[0167] Medium risk: 50% of the limit value ≤ PPV < 85% of the limit value;

[0168] High risk: PPV ≥ 85% of the limit value;

[0169] Early warning information can also be provided. For example, output three - level early warning signals:

[0170] Yellow early warning: The predicted vibration velocity reaches 70% of the limit value;

[0171] Orange early warning: The predicted vibration velocity reaches 85% of the limit value;

[0172] Red early warning: The predicted vibration velocity reaches 100% of the limit value.

[0173] The present invention provides an embodiment, which is: blasting vibration monitoring and prediction for a certain high-speed railway tunnel project. The tunnel length is 12.5 km, the rock mass grade is grade III (medium integrity), and the fracture density is 12 fractures / m 3 , groundwater condition: humid, sensitive target: there is a Ming Dynasty ancient building 80 m away from the tunnel axis (allowable vibration velocity ≤ 3 mm / s);

[0174] When arranging sensors, a main control node is arranged every 50 m, 4 dual-redundancy monitoring terminals are arranged around the ancient building, and sensors are arranged on the periphery of the main control node. Finally, 120 vibration sensors, 80 acoustic emission sensors, and 60 stress sensors are arranged;

[0175] Based on a comparison between the traditional method (Sadovsky formula) and the method described in the present invention, the superiority of the present invention is judged. The comparison results are shown in Table 1;

[0176] Table 1 Comparison result table

[0177]

[0178]

[0179] As can be seen from Table 1, the prediction error of the method of the present invention is significantly lower than that of the traditional method, and it can accurately predict the vibration influence range and trigger an early warning in time.

[0180] In this specification, each embodiment is described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other.

[0181] In this article, specific examples are used to elaborate on the principle and implementation manner of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those of ordinary skill in the art, according to the idea of the present invention, there will be changes in the specific implementation manner and application scope. To sum up, the content of this specification should not be construed as a limitation to the present invention.

Claims

1. A method for monitoring blasting vibration and predicting the influence range of a long tunnel in an environmentally sensitive area, characterized in that Including: Step 1: Construct a three-dimensional dynamic monitoring network; Step 2: Collect multi-source data based on the three-dimensional dynamic monitoring network and preprocess it to obtain a blasting vibration feature matrix; Step 3: Build a prediction model based on an improved LSTM model, and train the prediction model based on a preset blasting vibration feature matrix and its corresponding blasting parameters and geological conditions to obtain a trained prediction model; Step 4: Input the blasting vibration feature matrix to be detected and its corresponding blasting parameters and geological conditions into the trained prediction model to obtain a prediction result.

2. The method according to claim 1, wherein In Step 1, constructing a three-dimensional dynamic monitoring network specifically includes: Arranging monitoring terminals at regular intervals along the tunnel axis based on a drone cluster, setting double-redundant monitoring terminals around sensitive targets, and realizing real-time data transmission based on LoRa and 5G dual-mode communication. Among them, the monitoring terminals are equipped with vibration sensors, acoustic emission sensors, and stress sensors.

3. The method according to claim 2, wherein In Step 2, collecting multi-source data based on the three-dimensional dynamic monitoring network and preprocessing it to obtain a blasting vibration feature matrix specifically includes: Step 201: Collect the original vibration signal based on the vibration sensor, collect the original acoustic emission signal based on the acoustic emission sensor, and collect the original stress signal based on the stress sensor; Step 202: Perform signal denoising processing on the collected original vibration signal, original acoustic emission signal, and original stress signal; Step 203: Perform data normalization processing and time alignment processing on the vibration signal, acoustic emission signal, and stress signal after signal denoising processing; Step 204: Extract the features of the processed vibration signal, acoustic emission signal, and stress signal, and construct a blasting vibration feature matrix.

4. The method according to claim 3, wherein In Step 202, performing signal denoising processing on the collected original vibration signal, original acoustic emission signal, and original stress signal specifically includes: Performing denoising processing on the original vibration signal, original acoustic emission signal, and original stress signal based on an improved wavelet threshold denoising algorithm. The specific process is as follows: Discretize the original vibration signal, original acoustic emission signal, and original stress signal, perform 4-layer decomposition on the wavelet to obtain the approximate component and detail component of each layer of decomposition and the amplitude of the wavelet detail signal of that layer; Calculate the noise standard deviation and adaptive threshold of each layer of signal according to the detail signal coefficients of each layer of signal; Use the threshold quantization function to process the signal coefficients of each layer to complete the threshold denoising at each scale; Perform wavelet inverse transformation on the processed components of each layer to obtain the denoised components and complete wavelet reconstruction.

5. The method according to claim 3, wherein In Step 204, extracting the features of the processed vibration signal, acoustic emission signal, and stress signal and constructing a blasting vibration feature matrix specifically includes: For the vibration signal, extract its vibration features, including peak particle vibration velocity, main vibration frequency, and vibration duration; For the acoustic emission signal, extract its acoustic emission features, including acoustic emission event count, acoustic emission energy, and acoustic emission amplitude; For the stress signal, extract its stress features, including peak stress, stress change rate, and stress recovery time; Based on the vibration features, acoustic emission features, and stress features, construct a blasting vibration feature matrix F as: F = [PPV, main vibration frequency, duration, acoustic emission event count, acoustic emission energy, acoustic emission amplitude, peak stress, stress change rate, stress recovery time].

6. The method according to claim 5, wherein In step 3, a prediction model is built based on the improved LSTM model, and the prediction model is trained based on the preset blasting vibration feature matrix and its corresponding blasting parameters and geological conditions to obtain the trained prediction model. Specifically: Build a prediction model based on the LSTM-SA model, where the LSTM-SA model consists of a Self-Attention calculation module, an LSTM module, and a fully connected neural network; During its training process, the SGDR learning rate adjustment algorithm is used for training.

7. The method according to claim 6, wherein The blasting parameters include charge amount, burden, blasting method, and delay time, and the geological conditions include rock mass grade, fracture density, rock mass wave velocity, and groundwater condition.

Images