Blasting vibration monitoring data processing method based on neural network and kernel ridge regression

Abstract

The invention discloses a blasting vibration monitoring data processing method based on a neural network and kernel ridge regression, and relates to the technical field of engineering blasting. In order to solve the problem that a test result is inaccurate due to movement of a blasting source in on-line monitoring of field blasting vibration, a kernel ridge regression prediction model and a BP neural network are combined together, and a blasting vibration on-line monitoring data processing method based on the BP neural network and kernel ridge regression BP-KRR is provided. The method is used for correcting mine field blasting vibration online monitoring data. According to the method, a large amount of vibration data is generated as a data set of a training model by carrying out an impact test in a laboratory, model parameters are corrected, and blasting vibration waveform data are trained and corrected based on an artificial neural network BP model and a kernel ridge regression model, so that a correction model about distortion data caused by sensor azimuth change is obtained; the model can improve the on-site use flexibility of the blasting vibration meter and greatly improve the practical application capability of the blasting vibration meter.

Description

Technical Field

[0001] The invention belongs to the technical field of engineering blasting, and relates to a blasting vibration monitoring data processing method based on neural network and kernel ridge regression. Background Art

[0002] In the process of metal mining, rock drilling and blasting are the main way to break rocks. Therefore, frequent blasting is required in the daily production process of the mine. However, the disturbance caused by frequent blasting will induce cumulative damage to the open-pit slopes, underground pillars and roofs. The further development of this damage will induce disasters such as slope landslides, roof falls and empty area collapse, affecting the safe production of the mine. In order to evaluate this impact, the "Safety Regulations for Blasting" GB6722-2014 stipulates the blasting vibration measurement method and data processing.

[0003] Conventional blasting vibration meters have two core components, namely, blasting vibration recorder and vibration velocity sensor. The vibration velocity sensor has three channels, namely X, Y, and Z. In daily use, the X direction of the vibration velocity sensor of the device is aligned with the blasting vibration source as the benchmark for monitoring vibration, and the waveform received in the X direction is used for data analysis. At present, when the device is used for monitoring in mines, the equipment often moves with the blasting blasthole, but this vibration measurement method requires the tester to stay on site at the mine for testing, which does not meet the conditions for long-term monitoring. In order to monitor the impact of frequent blasting vibration on slopes, pillars, etc. for a long time, it is necessary to install long-term online monitoring equipment. Online monitoring requires that the blasting vibration meter be installed in a fixed position, and the direction of the vibration velocity sensor cannot be adjusted with the movement of the blasting source, resulting in the test data not meeting the requirements of the "Blasting Safety Regulations" and cannot be directly used for blasting vibration analysis. Therefore, it is necessary to propose a data processing method to correct the test data.

[0004] The existing technology for vibration monitoring at blasting sites mainly uses blasting vibrometers, which have two main components: a blasting vibration recorder and a vibration velocity sensor. At present, in order to address the defect that the vibration velocity sensor of the blasting vibrometer needs to be installed in a fixed position and the direction of the vibration velocity sensor cannot be adjusted as the blast source moves, a technology for remotely controlling the vibration velocity sensor has been proposed. This technology aims to achieve the purpose of remote control by modifying the structure of the sensor, adding a support plate and a movable base to the bottom of the vibration velocity sensor, and installing a remote drive on the support plate. However, this technology has the following disadvantages:

[0005] 1. Modifying equipment will not only increase costs, but also make it impossible to analyze the errors generated by the equipment;

[0006] 2. Although the purpose of remote control has been achieved, personnel are still required to operate the equipment every time a blast is performed, and the requirement of long-term online monitoring cannot be met. Summary of the invention

[0007] In view of the shortcomings of the prior art, the present invention combines the kernel ridge regression prediction model with the BP neural network to propose a blasting vibration online monitoring data processing method based on the BP neural network and the kernel ridge regression BP-KRR for correcting the online monitoring data of blasting vibration at the mine site. The method generates a large amount of vibration data by conducting laboratory impact tests, which is used as the data set for training the model, correcting the model parameters, and training and correcting the blasting vibration waveform data based on the artificial neural network BP model and the kernel ridge regression model, and deriving a correction model for the distorted data caused by the change of the sensor orientation. The model can improve the flexibility of the blasting vibration meter in the field and greatly improve its practical application ability.

[0008] To achieve the above purpose, the present invention proposes a blasting vibration monitoring data processing method based on neural network and kernel ridge regression, the steps are as follows:

[0009] Step 1: Obtain vibration waveforms under different vibration conditions and construct a data set;

[0010] Specifically, first, several earthquake sources are determined and different monitoring radius is set for each earthquake source. Then, monitoring points with the same monitoring radius but different angles are arranged around each earthquake source. Vibration waveforms of monitoring point tests at different angles under each earthquake source are obtained by using vibration velocity sensors at the monitoring points, and automatically saved in the blasting vibration recorder, thereby obtaining a data set including several vibration waveforms. The monitoring radius is the distance from the earthquake source to the monitoring point. Each sample in the data set is a vibration waveform and its corresponding distance from the earthquake source to the monitoring point, angle and energy of the earthquake source.

[0011] Step 2: Export the vibration waveform in the blasting vibration recorder, and preprocess the vibration waveform to obtain a preprocessed data set;

[0012] The preprocessing is specifically as follows: performing spectrum analysis on the vibration waveform by Fourier transform FFT to extract the main frequency and peak vibration velocity of the vibration waveform at each monitoring point; each sample in the preprocessed data set is a vibration waveform and its corresponding distance from the source to the monitoring point, angle, energy of the source, main frequency and peak vibration velocity of the vibration waveform;

[0013] Step 3: Divide the preprocessed data set into training set and test set according to the set ratio;

[0014] Step 4: construct a BP-KRR model; the BP-KRR model includes a BP neural network model and a kernel ridge regression KRR model connected in sequence;

[0015] Step 5: Use the training set to train the BP-KRR model to obtain a trained BP-KRR model;

[0016] Step 5.1: Input the training set into the BP neural network model for model training, and obtain the predicted main frequency and vibration speed output by the BP neural network model;

[0017] Step 5.2: Import the predicted main frequency and vibration speed into the kernel ridge regression KRR model for training, correct the predicted main frequency and vibration speed, and obtain the BP-KRR correction value;

[0018] Step 5.3: Repeat steps 5.1 to 5.2 to perform the set rounds of training to obtain the trained BP-KRR model;

[0019] Step 6: Use the test set to evaluate and adjust the trained BP-KRR model to obtain the final BP-KRR model;

[0020] Specifically: the test set is input into the trained BP-KRR model, the BP-KRR correction value is output, and the accuracy of the BP-KRR correction value is calculated. According to the accuracy result, the iteration function of the BP-KRR model is adjusted until the accuracy result reaches the set threshold, and the final BP-KRR model is obtained;

[0021] Step 7: Use the final BP-KRR model to process the subsequent monitoring data to obtain correct long-term online monitoring data; the long-term online monitoring data includes vibration speed and frequency;

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] Aiming at the problem of monitoring data distortion caused by blasting source movement in long-term online monitoring, a blasting vibration monitoring data processing method based on neural network and kernel ridge regression is proposed. This method has the advantages of BP neural network and kernel ridge regression prediction model, has low requirements on the quality of initial collected data, can be applied to the complex testing environment of mines, and can correct the long-term online monitoring data of blasting vibration measurement system with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Arrangement diagram of vibration velocity sensor in an embodiment of the present invention;

[0025] Figure 2 BP neural network structure diagram in the embodiment of the present invention;

[0026] Figure 3 BP neural network and kernel ridge regression prediction model in the embodiment of the present invention;

[0027] Figure 4Comparison chart of BP+KRR predicted value and actual value in the embodiment of the present invention;

[0028] Among them, (a) is the comparison diagram of vibration velocity model; (b) is the comparison diagram of main frequency model. DETAILED DESCRIPTION

[0029] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0030] The present invention proposes a blasting vibration monitoring data processing method based on neural network and kernel ridge regression to correct the monitoring data distortion problem. The steps are as follows:

[0031] Step 1: Obtain vibration waveforms under different vibration conditions and construct a data set;

[0032] Specifically, first, several earthquake sources are determined and different monitoring radius is set for each earthquake source. Then, monitoring points with the same monitoring radius but different angles are arranged around each earthquake source. Vibration waveforms of monitoring point tests at different angles under each earthquake source are obtained by using vibration velocity sensors at the monitoring points, and automatically saved in the blasting vibration recorder, thereby obtaining a data set including several vibration waveforms. The monitoring radius is the distance from the earthquake source to the monitoring point. Each sample in the data set is a vibration waveform and its corresponding distance from the earthquake source to the monitoring point, angle and energy of the earthquake source.

[0033] In this embodiment, eight blasting vibration meters of the same model, a common tape measure, an angle measuring instrument and a 4.14 kg weight are prepared; Figure 1 The vibration velocity sensor is arranged as shown, and a heavy hammer is placed at a height of 1m above the ground and falls freely, hitting the ground to generate vibration waves; the landing point of the heavy hammer is used as the center of the circle, and a circle with a monitoring radius of 1m is divided. Eight blasting vibration meters are arranged at 0°, 45°, 90°, 135°, 180°, 225°, 270° and 315° on the circumference of the circle, and they are numbered 1-8 respectively. The falling height of the heavy hammer ranges from 1m to 0.1m, with an interval of 0.1m. The heavy hammer is repeated for 3 times at each height interval, and the vibration data is recorded. In the experiment, it is required that the x-axis of the vibration velocity sensor points to the vibration center, and the vibration data is used to verify the uniformity of the site.

[0034] After verifying the uniformity of the site, the angles between the x-axis and the vibration center of the eight vibration velocity sensors were changed to 0°, 45°, 90°, 135°, 180°, 225°, 270° and 315° respectively.

[0035] When the angles of the eight vibration velocity sensors remain unchanged, the distance between the vibration velocity sensor and the landing point of the heavy hammer is changed ( Figure 1), 11 concentric circles with monitoring radii of 3m, 2.8m, 2.6m, 2.4m, 2.2m, 2.0m, 1.8m, 1.6m, 1.4m, 1.2m, and 1.0m were set respectively. The data collected by each concentric circle with different monitoring radii were recorded as a group, for a total of 11 groups of data;

[0036] According to the monitoring radius set above, the falling height of the weight was changed at each monitoring radius, from 2m to 1m, with an interval of 0.1m. The weight was dropped 3 times at each height interval, and the vibration data was recorded.

[0037] Step 2: Perform waveform analysis preprocessing on the collected data. After screening the collected effective vibration waveform data, perform spectrum analysis on the waveform through Fourier transform (FFT) to extract the main frequency and peak vibration velocity of the vibration waveform of each monitoring point.

[0038] Step 3: Divide the preprocessed data set into a training set and a test set with a ratio of 80% and 20% respectively. The divided data set is used to combine the algorithm to prepare for the subsequent training of the BP neural network model, BP-KRR model and model accuracy evaluation;

[0039] Step 4: construct a BP-KRR model; the BP-KRR model includes a BP neural network model and a kernel ridge regression KRR model connected in sequence;

[0040] Step 5: Use the training set to train the BP-KRR model to obtain a trained BP-KRR model;

[0041] Step 5.1: Input the training set into the BP neural network model for model training, and obtain the predicted main frequency and vibration speed output by the BP neural network model;

[0042] like Figure 2 , the impact energy generated by the free fall of the heavy hammer at different heights is used as the input x1 of the input layer. The angle between the x-axis of the vibration velocity sensor and the vibration center is set as the input x2 of the input layer, the distance between the landing center and the sensor is set as the input x3 of the input layer, and the main frequency and peak vibration velocity of the waveform are set as the input x4 and x5 of the input layer respectively. The number of hidden layers of the BP neural network is set to 2, and the output layer outputs the main frequency y1 and the vibration velocity y2.

[0043] Step 5.2: If Figure 3 As shown, the predicted main frequency and vibration speed are introduced into the kernel ridge regression KRR model for training, and the predicted main frequency and vibration speed are corrected to obtain the BP-KRR correction value;

[0044] Step 5.3: Repeat steps 5.1 to 5.2 to perform the set rounds of training to obtain the trained BP-KRR model;

[0045] Step 6: Use the test set to evaluate and adjust the trained BP-KRR model to obtain the final BP-KRR model;

[0046] In this implementation, 80% of the preprocessed data is used as a training set in proportion and imported into the BP neural network model for model training. The predicted value with the smallest error with the actual value is obtained by adjusting the model parameters; the obtained predicted value is imported into the kernel ridge regression model for model training again, and the prediction accuracy is calculated using the test set to evaluate the performance of the prediction model; according to the evaluation results, the model structure of the algorithm is adjusted until the accuracy reaches more than 90%.

[0047] Step 7: Use the final BP-KRR model to process the subsequent monitoring data to obtain correct long-term online monitoring data;

[0048] like Figure 4 As shown, by comparing the BP neural network prediction value, the BP-KRR prediction value and the actual value, the BP-KRR prediction value of the present invention is closer to the actual value.

Claims

1. A method for processing blasting vibration monitoring data based on neural network and kernel ridge regression, characterized in that: The steps include: Step 1: Obtain vibration waveforms under different vibration conditions and construct a data set; Step 2: Export the vibration waveform in the blasting vibration recorder, and preprocess the vibration waveform to obtain a preprocessed data set; Step 3: Divide the preprocessed data set into training set and test set according to the set ratio; Step 4: Construct BP-KRR model; Step 5: Use the training set to train the BP-KRR model to obtain a trained BP-KRR model; Step 6: Use the test set to evaluate and adjust the trained BP-KRR model to obtain the final BP-KRR model; Step 7: Use the final BP-KRR model to process the subsequent monitoring data to obtain correct long-term online monitoring data.

2. The method for processing blasting vibration monitoring data based on neural network and kernel ridge regression according to claim 1 is characterized in that: Step 1 is specifically as follows: first, determine several earthquake sources and set a different monitoring radius for each earthquake source, then arrange monitoring points with the same monitoring radius but different angles around each earthquake source, and use vibration velocity sensors at the monitoring points to obtain vibration waveforms of monitoring points tested at different angles for each earthquake source, and automatically save them in the blasting vibration recorder, thereby obtaining a data set including several vibration waveforms; the monitoring radius is the distance from the earthquake source to the monitoring point.

3. The method for processing blasting vibration monitoring data based on neural network and kernel ridge regression according to claim 1 is characterized in that: Each sample in the data set in step 1 is a vibration waveform and its corresponding distance, angle and energy of the earthquake source from the earthquake source to the monitoring point.

4. The method for processing blasting vibration monitoring data based on neural network and kernel ridge regression according to claim 1, characterized in that: The preprocessing described in step 2 is specifically: performing spectral analysis on the vibration waveform through Fourier transform FFT, extracting the main frequency and peak vibration velocity of the vibration waveform of each monitoring point; each sample in the preprocessed data set is a vibration waveform and its corresponding distance from the source to the monitoring point, angle, energy of the source, main frequency and peak vibration velocity of the vibration waveform.

5. The method for processing blasting vibration monitoring data based on neural network and kernel ridge regression according to claim 1, characterized in that: The BP-KRR model in step 4 includes a BP neural network model and a kernel ridge regression KRR model connected in sequence.

6. The method for processing blasting vibration monitoring data based on neural network and kernel ridge regression according to claim 1, characterized in that: Step 5 specifically includes: Step 5.1: Input the training set into the BP neural network model for model training, and obtain the predicted main frequency and vibration speed output by the BP neural network model; Step 5.2: Import the predicted main frequency and vibration speed into the kernel ridge regression KRR model for training, correct the predicted main frequency and vibration speed, and obtain the BP-KRR correction value; Step 5.3: Repeat steps 5.1 to 5.2 to perform the set rounds of training to obtain the trained BP-KRR model.

7. The method for processing blasting vibration monitoring data based on neural network and kernel ridge regression according to claim 1, characterized in that: Step 6 is as follows: input the test set into the trained BP-KRR model, output the BP-KRR correction value and calculate the accuracy of the BP-KRR correction value, and adjust the iteration function of the BP-KRR model according to the accuracy result until the accuracy result reaches the set threshold to obtain the final BP-KRR model.

8. The method for processing blasting vibration monitoring data based on neural network and kernel ridge regression according to claim 1, characterized in that: The long-term online monitoring data in step 7 includes vibration velocity and frequency.

Images