WO2022121139A1

WO2022121139A1 - Coal/gangue recognition method in top coal caving process based on multi-sensor information fusion

Info

Publication number: WO2022121139A1
Application number: PCT/CN2021/080195
Authority: WO
Inventors: 司垒; 王忠宾; 谭超; 闫海峰; 朱真才; 刘送永; 刘新华; 李嘉豪; 江红祥
Original assignee: 中国矿业大学
Priority date: 2020-12-11
Filing date: 2021-03-11
Publication date: 2022-06-16
Also published as: CN112465063A; CN112465063B

Abstract

A coal/gangue recognition method in a top coal caving process based on multi-sensor information fusion. By means of the combination of a BP neural network and a support vector machine, the BP neural network learns screened vibration signals, a trained BP neural network classification model can accurately recognize and classify the vibration signals generated by a coal/gangue impacting on a tail beam of a hydraulic support for top coal caving, the learning training of the support vector machine on screened sound signals can accurately and efficiently recognize and classify the sound signals generated by the coal/gangue impacting on the tail beam of the hydraulic support for top coal caving, a D-S evidence performs decision-level fusion on predicted structures to improve the accuracy and credibility of coal/gangue recognition, and a spectrum recognition apparatus can monitor the coal/gangue on a scraper in real time, which is an important basis for controlling the opening or closing of an opening for top coal caving. Since two recognition approaches are provided, the coal/gangue recognition precision is further improved; and the method not only can adapt to working conditions having good visibility, but also can adapt to working conditions having poor visibility.

Description

A method for identifying coal gangue in top coal caving process based on multi-sensor information fusion

technical field

The invention relates to a coal gangue identification method, in particular to a coal gangue identification method in a top coal caving process based on multi-sensor information fusion, and belongs to the technical field of coal gangue identification.

Background technique

In my country, the reserves of extra-thick coal seams are abundant and the fully mechanized caving mining method is mainly adopted. The realization of safe and efficient mining of extra-thick coal seams is of great significance to ensure the continuous supply of coal in my country. At present, the fully mechanized caving mining still adopts manual coal caving. Due to the large dust and harsh conditions on the coal mining face, it often brings safety problems to the on-site operators, and it is difficult to accurately determine the degree of top coal caving by manual operation, which inevitably leads to coal caving. Over-discharge status and under-discharge status of the process. The over-discharge condition will release a large amount of charcoal stones from the roof, resulting in a decrease in coal quality and an increase in transportation and washing costs; under-discharge conditions will result in loss of coal and a reduction in the recovery rate. Therefore, it is urgent to realize the automation of coal drawing process, and the key technology that must be realized in coal gangue identification.

Most of the coal gangue identification methods currently studied are passive identification, that is, identification is carried out according to the existing differences in chemical composition, physical properties, appearance and color of coal and gangue. The serious interference of the equipment has led to the unsatisfactory experimental results of the previous coal gangue identification methods in the field. Therefore, it is necessary to explore new and reliable automatic coal gangue identification methods.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for identifying coal gangue in the process of top coal caving based on multi-sensor information fusion, which can solve the problem of low accuracy of manual recognition of coal gangue caused by harsh environments such as dusty, humid, and dim working faces, Keep workers away from the hydraulic support operation area and reduce the labor intensity of workers.

In order to achieve the above purpose, the present invention provides a method for identifying coal gangue in a top coal caving process based on multi-sensor information fusion, comprising the following steps:

Step 1: Install the coal gangue identification device and the spectral identification device, and install the coal gangue identification device at the tail beam of the hydraulic support. The sensor, signal collector, microcomputer and intrinsically safe power supply, audio sensor and vibration sensor are respectively connected with the signal input end of the signal collector. The signal collector transmits the collected signal to the microcomputer through the network cable for processing and analysis. The computer is connected with the hydraulic support controller. The hydraulic support controller makes corresponding control actions according to the coal gangue identification results output by the microcomputer. The signal output end of the hydraulic support controller is connected to the signal collector. When the hydraulic support controller sends out coal discharge When the control command is received, the signal collector and the microcomputer start to collect and process the signal, thereby reducing the energy consumption and prolonging the service life of the device;

The spectral identification device includes an integrated probe installed at the lower part of the hydraulic support and obliquely above the rear scraper. The integrated probe is provided with a halogen light source and a collimating lens. The signal output end of the integrated probe is connected to the laser indicating light source, The signal input end of the spectrometer and the signal output end of the spectrometer are connected to the microcomputer;

Step 2: The identification method of the coal gangue identification device is:

① Before the hydraulic support automatically discharges the coal, the start and stop actions of the coal discharge are manually controlled, and the time is recorded as T ₁ . The audio sensor, vibration sensor and signal collector are used to collect the corresponding sound signals and vibration signals, and the collected signals are collected. transfer to a microcomputer for processing and storage;

②The microcomputer marks the sound signal and vibration signal generated by coal or gangue discharge. If coal is discharged, it will be marked as 1, and if gangue is discharged, it will be marked as 0. At the same time, the marked sound signal or vibration signal will be marked every 1s. Record it as 1 sample, and perform signal decomposition, feature extraction and feature screening for the sound signal and vibration signal with a sampling time of 1 second;

③ The microcomputer stores the marked filtered features, and uses them as the initial sample set to train the two classifiers, the support vector machine and the BP neural network. When the test error of the support vector machine and the BP neural network is less than the set value When the threshold value is ε, the training is over. At this time, the manual control of the start and stop of coal drawing is stopped, and the hydraulic support automatically starts to draw coal, and the time is recorded as T ₂ ;

④After the automatic coal discharge of the hydraulic support starts, the microcomputer performs signal decomposition, feature extraction and feature screening for the sound signal and vibration signal in two adjacent sampling times (ie T ₂ +1 seconds and T ₂ +2 seconds), respectively. And input them to the support vector machine and BP neural network classifier trained in step 3 respectively, and 4 prediction results can be obtained in the samples collected at T ₂ +1 seconds and T ₂ +2 seconds;

⑤ The microcomputer uses the D-S evidence theory to fuse the four prediction results obtained in step ④ at the decision level, so as to obtain the final coal gangue identification result;

⑥ The microcomputer sends the coal gangue identification result to the hydraulic support controller. When the identification result is gangue, the hydraulic support controller sends the command to stop coal caving, the tail beam of the hydraulic support extends, and the coal caving action stops;

The identification method of the spectral identification device is:

①Adjust the inclination angle of the integrated probe, turn on the halogen light source to illuminate the moving coal gangue on the rear scraper, so that the halogen light source illuminates the middle position of the rear scraper;

② After the coal discharge starts, the coal and gangue above the hydraulic support slide down to the rear scraper under the swing of the tail beam of the support. The reflected signal of coal or gangue moving on the rear scraper, and through the branch end of the Y-shaped optical fiber, the collected reflected signal is transmitted to the spectrometer, and the microcomputer is used to analyze the spectral data in the spectrometer. The pattern matching is performed on the spectral data of the rear scraper, and the coal or gangue passing through the field of view of the collimating lens on the rear scraper is qualitatively analyzed. In the microcomputer, the type of coal gangue is assigned as: coal=0, gangue=1;

③ Through the continuous release of coal and gangue, the scraper is constantly alternating between coal and gangue, and the values of 0 and 1 are accumulated in the microcomputer. When 0+1+0+1+0+1...=x (x represents the average number of gangue layers in the coal drawing face), it means that the coal drawing is over, and the microcomputer sends an instruction to control the closing of the coal drawing port through the hydraulic support controller.

As a further improvement of the present invention, the steps of signal decomposition, feature extraction and feature screening in step 2 (2) are as follows:

A: Using empirical mode decomposition (EMD) to decompose the original sound signal and vibration signal separately, each original signal can obtain several eigenmode functions IMF, and the number of IMFs of the original sound signal is recorded as m, the original vibration signal The number of IMFs is denoted as n;

B: For each original signal, calculate the energy E _i , kurtosis κ _i of each eigenmode function IMF and the correlation coefficient ζ _i of the original signal, and normalize each parameter, denoted as CE _i , Cκ respectively _i , Cζ _i ;

C: Calculate the weighted score of each eigenmode function IMF ρ _i =α×CE _i +β×Cκ _i +γ×Cξ _i , where α+β+γ=1, select the p eigenvalues with the highest weighted scores Modal function IMF for subsequent feature extraction;

D: For each original quotation mark, calculate the normalized eigenenergy CE _i and kurtosis Cκ _i of the extracted p eigenmode functions IMF, and use them as the initial sample set for SVM and BP neural network.

As a further improvement of the present invention, the radial basis kernel function is selected as the kernel function of the support vector machine.

As a further improvement of the present invention, the training steps of SVM and BP neural network in step 2 are as follows:

A: Parameter setting, the kernel function parameters and error penalty factor in the support vector machine are determined by cross-validation method, the number of nodes in the input layer of the BP neural network is p, the number of nodes in the output layer is 1, and the number of nodes in the hidden layer is satisfied.

q is a constant between 0-10, and then the optimal number of nodes is determined by trial and error;

B _: Denote the number of samples in the time T1 _- T2 as M, randomly select 60% of the M samples as the training sample set of SVM and BP neural network, and the rest as the test sample set, and support the trained Vector machine and BP neural network model for testing;

C: When the test accuracy of SVM and BP neural network is less than the set threshold ε, the training ends.

As a further improvement of the present invention, the step that microcomputer utilizes D-S evidence theory to carry out decision-level fusion to 4 prediction results in step 2 is as follows:

A: Denote the SVM and BP neural network output results of the sound signal samples collected within T ₂ +1 second as a1 and b1, respectively, and the SVM and BP of the vibration signal samples collected within T ₂ +1 second The output results of the neural network are respectively denoted as c1 and d1; similarly, the output results of the sensing signal samples collected within T ₂ +2 seconds are a2, b2, c2 and d2.

B: Normalize the 4 output results in the same sampling time respectively:

a11=a1/(a1+b1+c1+d1),

b11=b1/(a1+b1+c1+d1),

c11=c1/(a1+b1+c1+d1),

d11=d1/(a1+b1+c1+d1),

a22=a2/(a2+b2+c2+d2),

b22=b2/(a2+b2+c2+d2),

c22=c2/(a2+b2+c2+d2),

d22=d2/(a2+b2+c2+d2);

C: In the DS evidence theory, the output results at T ₂ +1 and T ₂ +2 are set as two pieces of evidence m1 and m2, namely:

Normalization constant K=a11×a22+b11×b22+c11×c22+d11×d22

Then the output result of the support vector machine based on the sound signal is a11×a22/K,

The output result of the BP neural network based on the sound signal is b11×b22/K,

The output result of the support vector machine based on the vibration signal is c11×c22/K,

The output result of BP neural network based on vibration signal is d11×d22/K;

D: If max{a11×a22/K, b11×b22/K, c11×c22/K, d11×d22/K}>0.5, then the final recognition result is recorded as coal discharge; otherwise, the final recognition result is recorded as Let go.

As a further improvement of the present invention, the sampling frequency of the audio sensor set by the signal collector is 45KHz, the sampling frequency of the vibration sensor is 20KHz, the sound sensor is a capacitive sensor, and the vibration sensor is a voltage-type or current-type high-frequency sensor.

As a further improvement of the present invention, the bifurcated optical fiber adopts a Y-shaped optical fiber, the optical fiber merging section is connected to the collimating lens embedded in the geometric center of the integrated probe, and the branch ends of the bifurcated optical fiber are respectively connected to the laser indicating light source and the spectrometer. The range captured by the collimating lens is indicated and the angle of installation inclination of the integrated probe is assisted in adjusting.

Compared with the prior art, the present invention utilizes the fusion method of the BP neural network model and the support vector machine, and fully combines the BP neural network, which can realize the complex nonlinear mapping relationship between input and output, and can approximate any nonlinear function, classifying The speed is fast, and the support vector machine can automatically find those support vectors that have better distinguishing ability for classification. The constructed classifier can maximize the interval between classes and has the advantages of better promotion performance and better classification accuracy. , therefore, the present invention can improve the classification accuracy of the sound sampling samples and the vibration sampling samples; the present invention skillfully integrates the advantages of the BP neural network and the support vector machine through an organic combination, thereby not only improving the filtered sound characteristic signals and The accuracy of the vibration signal features, and then the test accuracy can be effectively improved. The invention makes full use of the BP neural network and the support vector machine to perform particularly well in the classification effect of the vibration signal and the sound signal, and has high classification accuracy and high classification accuracy. In addition to the characteristics of better promotion performance, it also has the characteristics of simple and effective implementation. The method of the invention is simple to implement and low in cost. The BP neural network and the support vector machine are used to learn and train the filtered vibration signal and sound signal samples respectively. The proposed BP neural network classification model and support vector machine classification model can both realize accurate and efficient identification and classification of the vibration signal and sound signal generated by the impact of coal gangue on the tail beam of the top coal caving hydraulic support. The decision-level fusion greatly improves the accuracy and reliability of coal gangue identification, and can effectively eliminate the uncertain factors of multi-source signals, further improving the accuracy of coal gangue identification; the spectral identification device of the present invention has high fusion The intelligent top coal caving technology of spectrum and coal seam structure, according to the acquired spectral signal, qualitative analysis of coal, gangue and roof, and the alternately compared with the coal seam structure of the measured coal and gangue, can real-time Monitoring the coal gangue on the scraper is an important basis for controlling the opening and closing of the top coal caving port; because the present invention sets up two coal gangue identification methods, the coal gangue identification accuracy is further improved, and the coal gangue identification method of the present invention not only It can adapt to the working conditions of the underground fully mechanized caving face with harsh working conditions and low visibility, and can also adapt to the working conditions with better visibility.

Description of drawings

Fig. 1 is the principle block diagram of coal gangue identification device and spectral identification device of the present invention;

Fig. 2 is the flow chart of the coal gangue identification method of the present invention;

FIG. 3 is a flow chart of spectral identification of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings.

As shown in Figure 1 and Figure 2, a method for identifying coal gangue in the process of top coal caving based on multi-sensor information fusion includes the following steps:

Step 1: Install the coal gangue identification device and the spectral identification device, and install the coal gangue identification device at the tail beam of the hydraulic support. Sensor, signal collector, microcomputer and intrinsically safe power supply, audio sensor and vibration sensor are respectively connected with the signal input end of the signal collector, and the audio sensor and vibration sensor are respectively used to collect coal gangue impact caving in the process of top coal caving. The sound signal and vibration signal generated by the tail beam of the coal hydraulic support, the signal collector transmits the collected signal to the microcomputer through the network cable for processing and analysis, and the microcomputer is connected with the hydraulic support controller. The output coal gangue identification result makes corresponding control actions, and the signal output end of the hydraulic support controller is connected to the signal collector;

Step 2: The identification method of the coal gangue identification device is:

②The microcomputer marks the sound signal and vibration signal generated by coal or gangue discharge. If coal is discharged, it will be marked as 1, and if gangue is discharged, it will be marked as 0. At the same time, the marked sound signal or vibration signal will be marked every 1s. Record it as 1 sample, and perform signal decomposition, feature extraction and feature screening for the sound signal and vibration signal with a sampling time of 1 second. The steps of signal decomposition, feature extraction and feature screening are as follows:

(3) The microcomputer stores the marked filtered features, and trains the two classifiers, the support vector machine and the BP neural network as the initial sample set, respectively. When the test error of the support vector machine and the BP neural network is less than the set value When the threshold value is ε, the training is over. At this time, the manual control of the start and stop of coal drawing is stopped, and the hydraulic support automatically starts to draw coal, and the time is recorded as T ₂ ; the training steps of the support vector machine and the BP neural network are as follows:

C: When the test accuracy of the support vector machine and the BP neural network is less than the set threshold ε, the training ends;

The identification method of the spectral identification device is:

The kernel function of the support vector machine is a radial basis kernel function.

In the second step, the microcomputer uses the D-S evidence theory to perform decision-level fusion of the four prediction results as follows:

B: Normalize the 4 output results in the same sampling time respectively:

a11=a1/(a1+b1+c1+d1),

b11=b1/(a1+b1+c1+d1),

c11=c1/(a1+b1+c1+d1),

d11=d1/(a1+b1+c1+d1),

a22=a2/(a2+b2+c2+d2),

b22=b2/(a2+b2+c2+d2),

c22=c2/(a2+b2+c2+d2),

d22=d2/(a2+b2+c2+d2);

Normalization constant K=a11×a22+b11×b22+c11×c22+d11×d22

The output result of BP neural network based on vibration signal is d11×d22/K;

The sampling frequency of the audio sensor set by the signal collector is 45KHz, the sampling frequency of the vibration sensor is 20KHz, the sound sensor is a capacitive sensor, and the vibration sensor is a voltage-type or current-type high-frequency sensor.

The bifurcated optical fiber adopts Y-shaped optical fiber. The combined section of the optical fiber is connected to the collimating lens embedded in the geometric center of the integrated probe. The branch ends of the bifurcated optical fiber are respectively connected to the laser indicating light source and the spectrometer. The laser indicating light source can be aligned with the range collected by the collimating lens. Provides instructions and assists in adjusting the mounting tilt angle of the integrated probe.

The invention effectively combines the coal gangue identification method and the spectral identification method. When in use, the microcomputer analyzes the sensing data and the spectral data respectively, and controls the hydraulic support to stop the coal discharging operation when either identification method reaches its respective conditions. Compared with the prior art, the coal gangue recognition accuracy is improved to a great extent, and the coal gangue recognition method of the present invention can not only adapt to the harsh working environment and low visibility of the fully mechanized caving face, but also adapt to Good visibility conditions.

Claims

A method for identifying coal gangue in a top coal caving process based on multi-sensor information fusion, characterized in that it includes the following steps:

Step 1: Install the coal gangue identification device and the spectral identification device, and install the coal gangue identification device at the tail beam of the hydraulic support. The sensor, signal collector, microcomputer and intrinsically safe power supply, audio sensor and vibration sensor are respectively connected with the signal input end of the signal collector. The signal collector transmits the collected signal to the microcomputer through the network cable for processing and analysis. The computer is connected with the hydraulic support controller, the hydraulic support controller makes corresponding control actions according to the coal gangue identification result output by the microcomputer, and the signal output end of the hydraulic support controller is connected with a signal collector;

The spectral identification device includes an integrated probe installed at the lower part of the hydraulic support and obliquely above the rear scraper. The integrated probe is provided with a halogen light source and a collimating lens. The signal output end of the integrated probe is connected to the laser indicating light source, The signal input end of the spectrometer and the signal output end of the spectrometer are connected to the microcomputer;

Step 2: The identification method of the coal gangue identification device is:

① Before the hydraulic support automatically discharges the coal, the start and stop actions of the coal discharge are manually controlled, and the time is recorded as T 1 . The audio sensor, vibration sensor and signal collector are used to collect the corresponding sound signals and vibration signals, and the collected signals are collected. transfer to a microcomputer for processing and storage;

②The microcomputer marks the sound signal and vibration signal generated by coal or gangue discharge respectively. If coal is discharged, it is marked as 1; if gangue is discharged, it is marked as 0. 1s is recorded as 1 sample, and signal decomposition, feature extraction and feature screening are performed on the sound signal and the vibration signal with a sampling time of 1 second respectively;

③ The microcomputer stores the marked filtered features, and uses them as the initial sample set to train the two classifiers, the support vector machine and the BP neural network. When the test error of the support vector machine and the BP neural network is less than the set value When the threshold value is ε, the training is over. At this time, the manual control of the start and stop of coal drawing is stopped, and the hydraulic support automatically starts to draw coal, and the time is recorded as T 2 ;

④After the automatic coal discharge of the hydraulic support starts, the microcomputer performs signal decomposition, feature extraction and feature screening for the sound signal and vibration signal in two adjacent sampling times, namely T 2 +1 seconds and T 2 +2 seconds, and Input to the support vector machine and BP neural network classifier trained in step 3 respectively, and 4 prediction results can be obtained from the samples collected at T 2 +1 seconds and T 2 +2 seconds;

⑤ The microcomputer uses the D-S evidence theory to fuse the four prediction results obtained in step ④ at the decision level, so as to obtain the final coal gangue identification result;

⑥ The microcomputer sends the coal gangue identification result to the hydraulic support controller. When the identification result is gangue, the hydraulic support controller sends the command to stop coal caving, the tail beam of the hydraulic support extends, and the coal caving action stops;

The identification method of the spectral identification device is:

①Adjust the inclination angle of the integrated probe, turn on the halogen light source to illuminate the moving coal gangue on the rear scraper, so that the halogen light source illuminates the middle position of the rear scraper;

② After the coal discharge starts, the coal and gangue above the hydraulic support slide down to the rear scraper under the swing of the tail beam of the support. The reflected signal of coal or gangue moving on the rear scraper, and through the branch end of the Y-shaped optical fiber, the collected reflected signal is transmitted to the spectrometer, and the microcomputer is used to analyze the spectral data in the spectrometer. The pattern matching is performed on the spectral data of the rear scraper, and the qualitative analysis is carried out on the coal or gangue that has passed the field of view of the collimating lens on the rear scraper.

③ Through the continuous release of coal and gangue, the scraper is constantly alternating between coal and gangue, and the values of 0 and 1 are accumulated in the microcomputer. When 0+1+0+1+0+1...=x When it means that the coal draw is over, the microcomputer sends an instruction to control the closing of the coal draw port through the hydraulic support controller.
A method for identifying coal gangue in a top coal caving process based on multi-sensor information fusion according to claim 1, wherein the steps of signal decomposition, feature extraction and feature screening in step 2 (2) are as follows:

A: Using empirical mode decomposition (EMD) to decompose the original sound signal and vibration signal separately, each original signal can obtain several eigenmode functions IMF, and the number of IMFs of the original sound signal is recorded as m, the original vibration signal The number of IMFs is denoted as n;

B: For each original signal, calculate the energy E i , kurtosis κ i of each eigenmode function IMF and the correlation coefficient ζ i of the original signal, and normalize each parameter, denoted as CE i , Cκ respectively i , Cζ i ;

C: Calculate the weighted score of each eigenmode function IMF ρ i =α×CE i +β×Cκ i +γ×Cξ i , where α+β+γ=1, select the p eigenvalues with the highest weighted scores Modal function IMF for subsequent feature extraction;

D: For each original quotation mark, calculate the normalized eigenenergy CE i and kurtosis Cκ i of the extracted p eigenmode functions IMF, and use them as the initial sample set for SVM and BP neural network.
The method for identifying coal gangue in a top coal caving process based on multi-sensor information fusion according to claim 1 or 2, wherein the kernel function of the support vector machine is a radial basis kernel function.
A kind of gangue identification method in the top coal caving process based on multi-sensing information fusion according to claim 3, it is characterized in that, in step 2, the training steps of support vector machine and BP neural network are as follows:

A: Parameter setting, the kernel function parameters and error penalty factor in the support vector machine are determined by cross-validation method, the number of nodes in the input layer of the BP neural network is p, the number of nodes in the output layer is 1, and the number of nodes in the hidden layer is satisfied.
q is a constant between 0-10, and then the optimal number of nodes is determined by trial and error;

B : Denote the number of samples in the time T1 - T2 as M, randomly select 60% of the M samples as the training sample set of SVM and BP neural network, and the rest as the test sample set, and support the trained Vector machine and BP neural network model for testing;

C: When the test accuracy of SVM and BP neural network is less than the set threshold ε, the training ends.
A method for identifying coal gangue in top coal caving process based on multi-sensor information fusion according to claim 4, characterized in that in step 2, the microcomputer uses D-S evidence theory to perform decision-level fusion of the four prediction results as follows:

A: Denote the SVM and BP neural network output results of the sound signal samples collected within T 2 +1 second as a1 and b1, respectively, and the SVM and BP of the vibration signal samples collected within T 2 +1 second The output results of the neural network are respectively denoted as c1 and d1; similarly, the output results of the sensing signal samples collected within T 2 +2 seconds are a2, b2, c2 and d2.

B: Normalize the 4 output results in the same sampling time respectively:

a11=a1/(a1+b1+c1+d1),

b11=b1/(a1+b1+c1+d1),

c11=c1/(a1+b1+c1+d1),

d11=d1/(a1+b1+c1+d1),

a22=a2/(a2+b2+c2+d2),

b22=b2/(a2+b2+c2+d2),

c22=c2/(a2+b2+c2+d2),

d22=d2/(a2+b2+c2+d2);

C: In the DS evidence theory, the output results at T 2 +1 and T 2 +2 are set as two pieces of evidence m1 and m2, namely:

Normalization constant K=a11×a22+b11×b22+c11×c22+d11×d22

Then the output result of the support vector machine based on the sound signal is a11×a22/K,

The output result of the BP neural network based on the sound signal is b11×b22/K,

The output result of the support vector machine based on the vibration signal is c11×c22/K,

The output result of BP neural network based on vibration signal is d11×d22/K;

D: If max{a11×a22/K, b11×b22/K, c11×c22/K, d11×d22/K}>0.5, then the final recognition result is recorded as coal discharge; otherwise, the final recognition result is recorded as Let go.
The method for identifying coal gangue in the process of top coal caving based on multi-sensor information fusion according to claim 4, wherein the sampling frequency of the audio sensor set by the signal collector is 45KHz, and the sampling frequency of the vibration sensor is 20KHz.
The method for identifying coal gangue in the process of top coal caving based on multi-sensor information fusion according to claim 4, wherein the sound sensor is a capacitive sensor, and the vibration sensor is a voltage-type or current-type high-frequency sensor.
The method for identifying coal gangue in the process of top coal caving based on multi-sensing information fusion according to claim 7, wherein the bifurcated optical fiber adopts Y-shaped optical fiber, and the optical fiber merging section and the geometric center of the integrated probe are embedded The collimating lens is connected, and the branch ends of the bifurcated optical fibers are respectively connected to the laser indicating light source and the spectrometer. The laser indicating light source can indicate the range collected by the collimating lens and assist in adjusting the installation inclination angle of the integrated probe.