Multi-source information fusion coal rock identification method

Abstract

The invention discloses a coal rock identification method based on multi-source information fusion, and relates to the technical field of intelligent coal mining, and the method comprises the following steps: in the cutting process of a digging and anchoring all-in-one machine, collecting a current signal of a cutting roller motor and sound and vibration signals of a cutting roller through a sensor, and drawing a corresponding one-dimensional oscillogram; and carrying out direct current removal and noise reduction processing on the one-dimensional data of the current, sound and vibration signals, and converting the one-dimensional signal of a one-dimensional oscillogram into an angle and a radius through a Grubm angle field so as to generate a two-dimensional feature map. Current, sound and vibration signals in the cutting process of the digging and anchoring all-in-one machine are collected, collected one-dimensional data are converted into two-dimensional data through a Grubb angle field, and finally a coal rock interface is recognized through a convolutional neural network algorithm based on an attention mechanism. A cutting basis is provided for coal rock cutting in the tunneling process of the tunneling and anchoring all-in-one machine, and intelligent development of coal mining is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent coal mining, and particularly relates to a coal and rock identification method based on multi-source information fusion. Background Art

[0002] The roadway driving operation is not only an indispensable prerequisite for coal mining, but also the core link of the preparation work for the mining face. The driving speed of the coal mine roadway is directly related to the overall coal mining efficiency. Although significant progress has been made in Chinese coal mine technology in recent years, promoting the rapid development of roadway driving technology and equipment, due to the complexity of the coal resource storage environment, the technical and equipment levels of roadway driving are still the key factors restricting the high-yield and high-efficiency of coal mines. In addition, the geological conditions of the roadway are diverse and complex, bringing considerable challenges to the driving operation;

[0003] Therefore, it is necessary to design a coal and rock identification method based on multi-source information fusion to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a coal and rock identification method based on multi-source information fusion to solve the problems raised in the above background art.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A coal and rock identification method based on multi-source information fusion, comprising the following steps:

[0007] Step 1, during the cutting process of the roadheader, collect the current signal of the cutting drum motor, as well as the sound and vibration signals of the cutting drum through sensors, and draw the corresponding one-dimensional waveform diagram;

[0008] Step 2, perform DC removal and noise reduction processing on the one-dimensional data of the current, sound, and vibration signals, and convert the one-dimensional signal of the one-dimensional waveform diagram into angles and radii through the Gram angle field, and then generate a two-dimensional feature map;

[0009] Step 3, construct a convolutional neural network architecture for coal and rock identification, and introduce an attention mechanism to optimize the convolutional neural network;

[0010] Step 4, perform feature extraction, fusion on the current signal of the cutting drum motor of the roadheader, as well as the sound and vibration signals of the cutting drum, and use this for model training, and use the trained model to identify the coal and rock interface.

[0011] A further improvement of the technical solution of the present invention lies in: in the above Step 1, the process of drawing the one-dimensional waveform diagram is:

[0012] During the cutting process of the roadheader-anchoring machine, current sensors, sound sensors, and vibration sensors are deployed to monitor the signal changes of the cutting drum motor and the cutting drum;

[0013] Connect the current sensor to the circuit system of the cutting drum motor to monitor the change of the current intensity of the motor during operation in real time, output the captured current data in the form of digital signals, and the signals are then recorded and used for subsequent analysis;

[0014] Install the sound sensor around the cutting drum to capture the sound generated during the cutting process. The sound signal includes the frequency, amplitude, and duration of the sound;

[0015] Install the vibration sensor on the cutting drum to monitor the vibration during the cutting process. The vibration data shows the force on the mechanical components and the direct feedback of the coal and rock hardness;

[0016] Use the data acquisition system to convert the signals captured by the sensors into digital format and draw the corresponding one-dimensional waveform diagram. The one-dimensional waveform diagram intuitively shows the changes of current, sound, and vibration signals over time, providing a basis for subsequent data processing and coal and rock identification.

[0017] A further improvement of the technical solution of the present invention lies in that: in the step 2, the generation process of the two-dimensional feature map is as follows:

[0018] Subtract its average value from the original signal for DC removal processing to remove the DC component in the signal and obtain the signal after DC removal;

[0019] Adopt filtering technology to remove the high-frequency components in the signal, thereby reducing the noise components in the signal, extract the filtered signal, and improve the signal quality;

[0020] Normalize the one-dimensional signal after DC removal and noise reduction processing, and normalize the one-dimensional signal data to between 0 and 1 to ensure the stability and consistency of the signal during subsequent processing;

[0021] Use polar coordinates to represent the normalized time series, with the time stamp as the radius and the signal value as the angle, so that each time point is mapped to a polar coordinate point;

[0022] Convert the one-dimensional signal into a two-dimensional image through the Gram angle field, calculate the angle difference between any two time points in the normalized time series, and use GASF and GADF to encode the calculated angle difference. Convert the encoded angle difference into a two-dimensional feature map. Each feature map retains the time relationship and numerical relationship in the original signal and is presented in the form of a two-dimensional image. Among them, GASF encodes the angle difference by calculating the cosine value, and GADF encodes the angle difference by calculating the sine value.

[0023] A further improvement of the technical solution of the present invention lies in that: the calculation expression for normalizing one-dimensional signal data to between 0 and 1 is:

[0024]

[0025] In the formula, x(t) is the vibration signal at the original time t, and x t is the signal at this moment after scaling, X is the time series, and min(X) and max(X) are the minimum and maximum values in X respectively;

[0026] The calculation expression for representing the normalized time series using polar coordinates is:

[0027]

[0028] In the formula, t i is the timestamp code at the time of this point, N is a constant factor, the interval [-1, 1] or [0, 1] is divided into N equal parts, r is the polar axis, retaining the relationship in time, and φ i is the polar angle, retaining the relationship in value;

[0029] The calculation expression for encoding the calculation angle difference using GASF and GADF is:

[0030]

[0031] In the formula, φ i is the angle value of the i-th sequence, GASF is the calculation cosine value to encode the angle difference, GADF is the calculation sine value to encode the angle difference, and for a vibration signal with a length of n, a numerical matrix with a size of n×n is obtained through GASF and GADF encoding.

[0032] A further improvement of the technical solution of the present invention lies in that: in step 3, the process of optimizing the convolutional neural network is as follows:

[0033] Design a convolutional neural network architecture, which is composed of three convolutional units, a flattening layer, and three fully connected layers. Each convolutional unit is composed of a single convolutional layer, a ReLU activation function, and a max pooling layer;

[0034] Introduce an attention mechanism to optimize the convolutional neural network. Introducing the attention mechanism is the CBAM module. The attention mechanism can help the network focus on the most important parts of the input data and improve the recognition accuracy. The introduction of the attention mechanism is divided into two modules: the channel attention module and the spatial attention module;

[0035] For the channel attention module, the spatial content of all channels is summarized by performing average pooling and max pooling operations to construct two independent spatial feature representations, labeled as Fc-avg and Fc-max respectively. Fc-avg and Fc-max are then input into a multi-layer perceptron network with a single hidden layer, which is responsible for assigning weight parameters to different channels, thereby reflecting the proportion of each channel in the overall feature importance. After the weighting process, by element-wise adding the two spatial feature representations and processing them through the Sigmoid activation function, the channel weights are normalized to the range of 0 to 1. The normalized weights are then multiplied with the input feature map channel by channel to generate the final channel attention-regulated features;

[0036] In the processing sequence of the spatial attention module, the output features obtained from the channel attention module are placed into the spatial attention module again as its input data. In the specific steps, the input features are average-pooled and max-pooled at the channel level based on their respective spatial positions. The two pooled feature maps are connected in parallel and then subjected to a convolution operation to obtain a new feature representation. Finally, the sigmoid activation function is used to normalize the spatial weights of the convolutional feature representation, obtaining a normalized spatial weight in the range of 0 to 1. The normalized spatial weight is multiplied with the original input feature map element by element to generate the final spatial attention features.

[0037] A further improvement of the technical solution of the present invention lies in that the operation formula of the channel attention is as follows:

[0038] M c (F) = σ(W1(W0(W cavg )) + W1(W0(W cavg )));

[0039] Where σ represents the sigmoid activation function, and W0 and W1 are the weights of the multi-layer perceptron;

[0040] The calculation process formula of the spatial attention is as follows:

[0041] M s (F) = σ(f(F Savg ; F Smax ));

[0042] Where σ represents the sigmoid activation function, and f represents the convolution operator.

[0043] A further improvement of the technical solution of the present invention lies in that in step 4, the process of identifying the coal-rock interface is as follows:

[0044] Extract the features for identifying the coal-rock interface from the current, sound, and vibration signals respectively;

[0045] The extraction of the current signal features of the cutting drum motor of the roadheader-anchoring machine altogether includes eight features, namely, average value, variance, energy, average amplitude, root mean square, root amplitude, effective value, and effective value entropy. The convolutional neural network is used for feature extraction to output the current signal feature map;

[0046] The extraction of the sound signal features of the cutting drum of the roadheader-anchoring machine altogether includes fifteen features, namely, average value, variance, energy, average amplitude, root mean square, root amplitude, effective value, effective value entropy, peak factor, shape parameter, skewness coefficient, impulse factor, valley factor, kurtosis, and kurtosis factor. The convolutional neural network is used for feature extraction to output the sound signal feature map;

[0047] The extraction of the vibration signal features of the cutting drum of the roadheader-anchoring machine altogether includes fifteen features, namely, average value, variance, energy, average amplitude, root mean square, root amplitude, effective value, effective value entropy, peak factor, shape parameter, skewness coefficient, impulse factor, valley factor, kurtosis, and kurtosis factor. The convolutional neural network is used for feature extraction to output the vibration signal feature map;

[0048] The extracted features are normalized, and the feature maps output by the three signals are fused in a channel superposition manner to form a new feature map. Through the fused feature vector, the working state of the cutting drum of the roadheader-anchoring machine and the characteristics of the coal-rock interface can be more comprehensively reflected;

[0049] Collect the current signal, sound signal, and vibration signal data of the cutting drum of the roadheader-anchoring machine under different coal-rock interfaces to obtain new coal-rock interface sample data and label the corresponding coal-rock interface types;

[0050] Use the trained model to extract and fuse the features of the new current signal, sound signal, and vibration signal, and input the fused feature vector into the model to obtain the recognition result of the coal-rock interface;

[0051] Adopt accuracy rate, recall rate, and F1-score indicators to comprehensively evaluate the performance of the model in the coal-rock interface recognition task, and use the feature importance map method to evaluate the contribution of each feature to the model performance, and then calculate the feature evaluation index to analyze the recognition feature accuracy;

[0052] Iteratively optimize the model according to the evaluation results, and further adjust the model structure and parameters to improve the recognition accuracy.

[0053] A further improvement of the technical solution of the present invention lies in that: the calculation expression of the feature evaluation index is:

[0054]

[0055] Among them, FEI is the Feature Evaluation Index, m is the total number of features, w j is the weight of the j-th feature, which reflects the relative importance of this feature in the overall evaluation, f j is the output value of the j-th feature, such as any numerical value like accuracy, recall rate, F1 score, etc. b is the benchmark value, which is used to set the standard threshold of performance and is the preset target performance value. s is the standard deviation, which is used to measure the dispersion degree of feature values and reflects the variability of data. The value range of FEI is between 0 and 1.

[0056] Due to the adoption of the above technical solution, the technical progress achieved by the present invention compared with the prior art is as follows:

[0057] 1. The present invention provides a coal-rock identification method for multi-source information fusion, which collects current, sound, and vibration signals during the cutting process of a roadheader-anchoring machine, converts the collected one-dimensional data into two-dimensional data through the Gram angle field, and finally identifies the coal-rock interface through a convolutional neural network algorithm based on the attention mechanism, providing a cutting basis for coal-rock cutting during the tunneling process of the roadheader-anchoring machine and promoting the intelligent development of coal mine mining.

[0058] 2. The present invention provides a coal-rock identification method for multi-source information fusion, which has the function of identifying the underground coal-rock interface, can provide detailed cutting data during the tunneling process of the roadheader-anchoring machine, thereby automatically adjusting the cutting parameters, optimizing the cutting trajectory, and performing cutting operations more precisely, avoiding unnecessary cutting and waste, thus improving the tunneling efficiency. With the improvement of the tunneling efficiency, the coal mining volume of the coal mine also increases correspondingly, further enhancing the economic benefits. It can also monitor the changes of the coal-rock interface in real time, timely discover potential safety hazards, thereby avoiding accidents, and can also help the roadheader-anchoring machine avoid hard gangue, reduce the wear and replacement cycle of pick teeth, reduce equipment failure rate, and improve the starting efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0060] Figure 1 is the method flow chart of the present invention;

[0061] Figure 2 is the one-dimensional signal diagram collected by the present invention;

[0062] Figure 3 is the two-dimensional conversion diagram of the present invention;

[0063] Figure 4Flow chart of Gram angle field of the present invention;

[0064] Figure 5 Architecture diagram of the convolutional neural network of the present invention;

[0065] Figure 6 Flow chart for identifying coal-rock interface of the present invention. Specific implementation manners

[0066] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, 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 some but not all of the embodiments of the present invention. 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.

[0067] Example 1, as Figures 1 to 5 shown, the present invention provides a coal-rock identification method for multi-source information fusion, including the following steps:

[0068] Step 1, during the cutting process of the roadheader, collect the current signal of the cutting drum motor, as well as the sound and vibration signals of the cutting drum through sensors, and draw the corresponding one-dimensional waveform diagram. During the cutting process of the roadheader, deploy current sensors, sound sensors and vibration sensors to monitor the signal changes of the cutting drum motor and the cutting drum. Connect the current sensor to the circuit system of the cutting drum motor to monitor the change of the current intensity of the motor during operation in real time, and output the captured current data in the form of digital signals. The signals are then recorded and used for subsequent analysis. Install the sound sensor around the cutting drum to capture the sound generated during the cutting process. The sound signal includes the frequency, amplitude and duration of the sound. Install the vibration sensor on the cutting drum to monitor the vibration during the cutting process. The vibration data shows the force on the mechanical components and the direct feedback of the coal-rock hardness. Use the data acquisition system to convert the signals captured by the sensors into digital format and draw the corresponding one-dimensional waveform diagram. The one-dimensional waveform diagram intuitively shows the changes of the current, sound and vibration signals over time, providing a basis for subsequent data processing and coal-rock identification;

[0069] Step 2, perform DC removal and noise reduction processing on the one-dimensional data of the current, sound and vibration signals, and convert the one-dimensional signals of the one-dimensional waveform diagram into angles and radii through the Gram angle field, and then generate a two-dimensional feature map. Subtract the average value from the original signal for DC removal processing to remove the DC component in the signal and obtain the signal after DC removal. The calculation expression for DC removal processing is:

[0070] Among them, x[n] is the original signal, and x dc [n] is the signal after removing the DC component, and N is the length of the signal;

[0071] By using filtering technology, the high-frequency components in the signal are removed, thereby reducing the noise components in the signal, extracting the filtered signal, and improving the quality of the signal. The calculation expression of the filtered signal is:

[0072] Among them, y[n] is the filtered signal, and h[k] is the impulse response of the filter;

[0073] The one-dimensional signal after DC removal and noise reduction is normalized, and the one-dimensional signal data is normalized to between 0 and 1 to ensure the stability and consistency of the signal during subsequent processing. The normalized time series is represented using polar coordinates, with the timestamp as the radius and the signal value as the angle, so that each time point is mapped to a polar coordinate point. The one-dimensional signal is converted into a two-dimensional image through the Gram angle field. The angle difference between any two time points in the normalized time series is calculated, and the calculated angle difference is encoded using GASF and GADF, and the encoded angle difference is converted into a two-dimensional feature map. Each feature map retains the time relationship and numerical relationship in the original signal and is presented in the form of a two-dimensional image. Among them, GASF encodes the angle difference by calculating the cosine value, and GADF encodes the angle difference by calculating the sine value;

[0074] Furthermore, the calculation expression for normalizing the one-dimensional signal data to between 0 and 1 is:

[0075]

[0076] In the formula, x(t) is the vibration signal at the original t moment, and x t is the signal at this moment after scaling, X is the time series, and min(X) and max(X) are the minimum and maximum values in X respectively;

[0077] The calculation expression for representing the normalized time series using polar coordinates is:

[0078]

[0079] In the formula, t i is the timestamp code at this point moment, N is a constant factor, the interval [-1, 1] or [0, 1] is divided into N equal parts, r is the polar axis, which retains the time relationship, and φ i is the polar angle, which retains the numerical relationship;

[0080] The calculation expression for encoding the calculated angle difference using GASF and GADF is:

[0081]

[0082] where φ i is the angular value of the i-th sequence, GASF is the coded angular difference for calculating the cosine value, and GADF is the coded angular difference for calculating the sine value. For a vibration signal of length n, a numerical matrix of size n×n is obtained through GASF and GADF coding;

[0083] Step 3: Construct a convolutional neural network architecture for coal and rock identification. Introduce an attention mechanism to optimize the convolutional neural network. Design the convolutional neural network architecture, which consists of three convolutional units, a flattening layer, and three fully connected layers. Each convolutional unit is composed of a single convolutional layer, a ReLU activation function, and a max pooling layer. Among them, the convolutional layer is used to extract the spatial features of the input data, the ReLU activation function is used to introduce non-linearity to help the network learn complex patterns, the max pooling layer is used to reduce the spatial dimension of the features, reduce the computational amount, and extract important features. The fully connected layer is used to map the learned features to the final output, which can effectively extract the features in the coal and rock signals and provide a basis for subsequent classification and identification. Introduce an attention mechanism to optimize the convolutional neural network. The introduced attention mechanism is the CBAM module. The attention mechanism can help the network focus on the most important part of the input data and improve the recognition accuracy. The introduction of the attention mechanism is divided into two modules: the channel attention module and the spatial attention module. For the channel attention module, the spatial content of the entire channel is summarized by performing average pooling and max pooling operations to construct two independent spatial feature representations, respectively labeled as Fc-avg and Fc-max. Fc-avg and Fc-max are then input into a multi-layer perceptron network with a single hidden layer. This multi-layer perceptron network is responsible for assigning weight parameters to different channels, thereby reflecting the proportion of each channel in the overall feature importance. After completing a weighting process, the two spatial feature representations are element-wise added and processed through the Sigmoid activation function to normalize the channel weights to the interval from 0 to 1. The normalized weights are then multiplied by the input feature map channel by channel to generate the final channel attention-regulated features. For the processing sequence of the spatial attention module, the output features obtained from the channel attention module are placed into the spatial attention module again as its input data. In the specific steps, the input features are averaged and max-pooled at the channel level based on their respective spatial positions. The two pooled feature maps are connected in parallel and passed through a convolutional operation to obtain a new feature representation. Finally, the sigmoid activation function is used to normalize the spatial weights of the convolutional feature representation to obtain the normalized spatial weights in the range from 0 to 1. The normalized spatial weights are multiplied element-wise with the original input feature map to generate the final spatial attention features;

[0084] Furthermore, the operation formula of channel attention is as follows:

[0085] M c (F) = σ(W1(W0(W cavg )) + W1(W0(W cavg )));

[0086] where σ represents the sigmoid activation function, and W0 and W1 are the weights of the multi-layer perceptron;

[0087] The calculation process formula of spatial attention is as follows:

[0088] M s (F) = σ(f(F Savg ; F Smax ));

[0089] where σ represents the sigmoid activation function, and f represents the convolution operator;

[0090] Step 4: Extract, fuse the current signal of the cutting drum motor of the roadheader-anchoring machine, as well as the sound and vibration signals of the cutting drum, and perform model training based on this, and use the trained model to identify the coal-rock interface.

[0091] Example 2, as Figure 6 shown, based on Example 1, the present invention provides a technical solution: Preferably, in Step 4, the process of identifying the coal-rock interface is as follows:

[0092] Extract the features for identifying the coal-rock interface from the current, sound, and vibration signals respectively. The extraction of the current signal features of the cutting drum motor of the roadheader-anchoring machine includes a total of eight features, namely, mean value, variance, energy, average amplitude, root mean square, root amplitude, effective value, and effective value entropy. Use a convolutional neural network for feature extraction and output the current signal feature map. The extraction of the sound signal features of the cutting drum of the roadheader-anchoring machine includes a total of fifteen features, namely, mean value, variance, energy, average amplitude, root mean square, root amplitude, effective value, effective value entropy, peak factor, shape parameter, skewness coefficient, impulse factor, valley factor, kurtosis, and kurtosis factor. Use a convolutional neural network for feature extraction and output the sound signal feature map. The extraction of the vibration signal features of the cutting drum of the roadheader-anchoring machine includes a total of fifteen features, namely, mean value, variance, energy, average amplitude, root mean square, root amplitude, effective value, effective value entropy, peak factor, shape parameter, skewness coefficient, impulse factor, valley factor, kurtosis, and kurtosis factor. Use a convolutional neural network for feature extraction and output the vibration signal feature map. Normalize the extracted features, fuse the feature maps output by the three signals in a channel-overlapping manner to form a new feature map, and through the fused feature vector, more comprehensively reflect the working state of the cutting drum of the roadheader-anchoring machine and the features of the coal-rock interface. Collect the current signal, sound signal, and vibration signal data of the cutting drum of the roadheader-anchoring machine under different coal-rock interfaces, obtain new coal-rock interface sample data, and label the corresponding coal-rock interface types. Use the trained model to perform feature extraction and fusion on the new current signal, sound signal, and vibration signal, input the fused feature vector into the model, and obtain the recognition result of the coal-rock interface. Use the accuracy rate, recall rate, and F1 score indicators to comprehensively evaluate the performance of the model in the coal-rock interface recognition task, and use the feature importance map method to evaluate the contribution of each feature to the model performance, and then calculate the feature evaluation index to analyze the accuracy of the recognized features. Iteratively optimize the model according to the evaluation results, and further adjust the model structure and parameters to improve the recognition accuracy;

[0093] Furthermore, the calculation expression of the feature evaluation index is as follows:

[0094]

[0095] Among them, FEI is the feature evaluation index, m is the total number of features, w j is the weight of the j-th feature, which reflects the relative importance of this feature in the overall evaluation, f jis the output value of the j-th feature, such as any numerical value like accuracy, recall rate, F1 score, etc. b is the baseline value, used to set the standard threshold for performance, which is the preset target performance value. s is the standard deviation, used to measure the dispersion of feature values and reflect the variability of data. The value of FEI ranges from 0 to 1. When FEI is close to 1, it indicates that the performance of the feature set is much higher than the baseline value b, that is, the feature set makes a great contribution to the performance of the model and the prediction accuracy of the model is high. When FEI is close to 0, it indicates that the performance of the feature set is close to or lower than the baseline value b, that is, the feature set makes a small contribution to the performance of the model and the prediction accuracy of the model is low. The weighted sum of features is converted into a value between 0 and 1 through an exponential function, where the radical and summation functions are used to comprehensively consider the performance of all features, and the baseline value b is used to set a reference point for performance, more comprehensively evaluate the performance of the feature set, and optimize and iterate the model accordingly.

[0096] As described above, it is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A coal and rock identification method based on multi-source information fusion, characterized in that: It includes the following steps: Step 1, during the cutting process of the roadheader-anchoring machine, collect the current signal of the cutting drum motor, as well as the sound and vibration signals of the cutting drum through sensors, and draw the corresponding one-dimensional waveform diagram; Step 2, perform DC removal and noise reduction processing on the one-dimensional data of the current, sound, and vibration signals, and convert the one-dimensional signal of the one-dimensional waveform diagram into angles and radii through the Gram angular field, and then generate a two-dimensional feature map; Step 3, construct a convolutional neural network architecture for coal and rock identification, and introduce an attention mechanism to optimize the convolutional neural network; Step 4, perform feature extraction, fusion on the current signal of the cutting drum motor of the roadheader-anchoring machine, as well as the sound and vibration signals of the cutting drum, and use the trained model to identify the coal-rock interface.

2. The coal and rock identification method based on multi-source information fusion according to claim 1, characterized in that: In the said Step 1, the process of drawing the one-dimensional waveform diagram is as follows: During the cutting process of the roadheader-anchoring machine, deploy current sensors, sound sensors, and vibration sensors to monitor the signal changes of the cutting drum motor and the cutting drum; Connect the current sensor to the circuit system of the cutting drum motor, and monitor the change of the current intensity of the motor during operation in real time, and output the captured current data in the form of digital signals; Install the sound sensor around the cutting drum to capture the sound generated during the cutting process. The sound signal includes the frequency, amplitude, and duration of the sound; Install the vibration sensor on the cutting drum to monitor the vibration during the cutting process. The vibration data shows the force on the mechanical components and the direct feedback of the coal-rock hardness; Use the data acquisition system to convert the signals captured by the sensors into digital format and draw the corresponding one-dimensional waveform diagram. The one-dimensional waveform diagram intuitively shows the changes of the current, sound, and vibration signals over time.

3. A coal and rock identification method based on multi-source information fusion according to claim 2, characterized in that: In the said Step 2, the process of generating the two-dimensional feature map is as follows: Subtract its average value from the original signal for DC removal processing to remove the DC component in the signal and obtain the signal after DC removal; Adopt filtering technology to remove the high-frequency components in the signal, and then reduce the noise components in the signal to extract the filtered signal; Perform normalization processing on the one-dimensional signal after DC removal and noise reduction, and normalize the one-dimensional signal data to between 0 and 1; Use polar coordinates to represent the normalized time series, with the time stamp as the radius and the signal value as the angle, so that each time point is mapped to a polar coordinate point; Convert the one-dimensional signal into a two-dimensional image through the Gram angular field, calculate the angle difference between any two time points in the normalized time series, and use GASF and GADF to encode the calculated angle difference, and convert the encoded angle difference into a two-dimensional feature map. Each feature map retains the time relationship and numerical relationship in the original signal and is presented in the form of a two-dimensional image. Among them, GASF encodes the angle difference by calculating the cosine value, and GADF encodes the angle difference by calculating the sine value.

4. A coal and rock identification method based on multi-source information fusion according to claim 3, characterized in that: The calculation expression for normalizing the one-dimensional signal data to between 0 and 1 is: where \(x(t)\) is the vibration signal at the original time \(t\), and \(x\) t t is the signal at this time after scaling, \(X\) is the time series, and \(\min(X)\) and \(\max(X)\) are the minimum and maximum values in \(X\) respectively; The calculation expression for using polar coordinates to represent the normalized time series is: where \(t\) i is the timestamp code at this point in time, \(N\) is a constant factor that divides the interval \([-1, 1]\) or \([0, 1]\) into \(N\) equal parts, \(r\) is the polar axis, and \(\varphi\) i is the polar angle; The calculation expression for using GASF and GADF to encode the calculated angle difference is: where φ i is the angular value of the i-th sequence, GASF is the coded angular difference for calculating the cosine value, and GADF is the coded angular difference for calculating the sine value.

5. A coal and rock identification method for multi-source information fusion according to claim 4, characterized in that: In step 3, the process of optimizing the convolutional neural network is as follows: Design the convolutional neural network architecture, which consists of three convolutional units, a flattening layer, and three fully connected layers. Each convolutional unit is composed of a single convolutional layer, a ReLU activation function, and a max pooling layer; Introduce an attention mechanism to optimize the convolutional neural network. Introducing the attention mechanism means the CBAM module, and the introduction of the attention mechanism is divided into two modules: the channel attention module and the spatial attention module; For the channel attention module, summarize the spatial content of all channels by performing average pooling and max pooling operations to construct two independent spatial feature representations, labeled as Fc-avg and Fc-max respectively. Fc-avg and Fc-max are then input into a multi-layer perceptron network with a single hidden layer. This multi-layer perceptron network is responsible for assigning weight parameters to different channels. After the weighting process, by element-wise adding the two spatial feature representations and processing them through the Sigmoid activation function, normalize the channel weights to the range of 0 to 1. The normalized weights are then multiplied with the input feature map channel by channel to generate the final channel attention regulation feature; In the process of processing the spatial attention module, the output feature obtained from the channel attention module is placed into the spatial attention module again as its input data. The input feature is averaged and max-pooled at the channel level based on their respective spatial positions. The two pooled feature maps are connected in parallel and passed through a convolutional operation to obtain a new feature representation. Finally, use the sigmoid activation function to normalize the spatial weights of the convolutional feature representation, obtaining a normalized spatial weight in the range of 0 to 1. Multiply the normalized spatial weight with the original input feature map element by element to generate the final spatial attention feature.

6. The coal and rock identification method with multi-source information fusion according to claim 5, characterized in that: The operation formula of the channel attention is as follows: M c (F) = σ(W1(W0(W cavg )) + W1(W0(W cavg ))); where σ represents the sigmoid activation function, and W0 and W1 are the weights of the multi-layer perceptron; The calculation process formula of the spatial attention is as follows: M s (F) = σ(f([F Savg ; F Smax )); where σ represents the sigmoid activation function, and f represents the convolutional operator.

7. A coal and rock identification method for multi-source information fusion according to claim 6, characterized in that: In step 4, the process of identifying the coal-rock interface is as follows: Extract the features for identifying the coal-rock interface from the current, sound, and vibration signals respectively; The extraction of the current signal features of the cutting drum motor of the roadheader-anchoring machine in total includes eight features, namely average value, variance, energy, average amplitude, root mean square, root amplitude, effective value, and effective value entropy. Use the convolutional neural network to extract features and output the current signal feature map; The extraction of the sound signal features of the cutting drum of the roadheader-anchoring machine in total includes fifteen features, namely average value, variance, energy, average amplitude, root mean square, root amplitude, effective value, effective value entropy, peak factor, shape parameter, skewness coefficient, pulse factor, valley factor, kurtosis, and kurtosis factor. Use the convolutional neural network to extract features and output the sound signal feature map; The extraction of the vibration signal features of the cutting drum of the roadheader-anchoring machine altogether includes fifteen features, namely, average value, variance, energy, average amplitude, root mean square, root amplitude, effective value, effective value entropy, peak factor, shape parameter, skewness coefficient, impulse factor, valley factor, kurtosis, kurtosis factor. The convolutional neural network is used for feature extraction to output the vibration signal feature map; The extracted features are normalized, and the feature maps output by the three signals are fused in a channel superposition manner to form a new feature map, and through the fused feature vector, the working state of the cutting drum of the roadheader-anchoring machine and the characteristics of the coal-rock interface are more comprehensively reflected; Collect the current signal, sound signal and vibration signal data of the cutting drum of the roadheader-anchoring machine under different coal-rock interfaces to obtain new coal-rock interface sample data and label the corresponding coal-rock interface types; Use the trained model to extract and fuse the features of the new current signal, sound signal and vibration signal, input the fused feature vector into the model, and obtain the recognition result of the coal-rock interface; Adopt accuracy rate, recall rate, and F1 score indicators to comprehensively evaluate the performance of the model in the coal-rock interface recognition task, and use the feature importance map method to evaluate the contribution of each feature to the model performance, and then calculate the feature evaluation index to analyze the recognition feature accuracy; Iteratively optimize the model according to the evaluation results, and further adjust the model structure and parameters.

8. A coal and rock identification method based on multi-source information fusion according to claim 7, characterized in that: The calculation expression of the said feature evaluation index is: Among them, FEI is the feature evaluation index, m is the total number of features, w j is the weight of the j-th feature, f j is the output value of the j-th feature, b is the reference value, which is a preset target performance value, s is the standard deviation, and the value range of FEI is between 0 and 1.

Images