Self-adaptive noise reduction method for vibration signals of gas extraction drilling machine based on multi-scale feature fusion

Abstract

The invention relates to the technical field of gas extraction drilling machine vibration signal processing, in particular to a gas extraction drilling machine vibration signal self-adaptive noise reduction method based on multi-scale feature fusion. The method comprises the steps of collecting a vibration signal, converting the vibration signal into a two-dimensional waveform image, recognizing a drilling working condition area, constructing a multi-scale objective function to optimize VMD parameters, decomposing the signal and screening a dominant IMF component for reconstruction. Through multi-objective optimization and multi-feature fusion strategies of energy distribution and transient impact characteristics, the problems that in a traditional method, the noise reduction effect is poor, and the coal rock character recognition precision is insufficient are solved, the waveform similarity, the power spectrum density coincidence degree and the correlation coefficient are remarkably improved, and reliable guarantee is provided for coal mine safety production.

Description

Technical Field

[0001] The present invention belongs to the technical field of mechanical vibration signal denoising processing, and specifically provides an adaptive denoising method for vibration signals of a gas extraction drilling rig based on multi-scale feature fusion. Background Art

[0002] Gas extraction is an important means of preventing and controlling coal mine gas disasters, developing coalbed methane resources, and reducing greenhouse gas emissions. Drilling and extraction is the primary technical method and core work content for achieving gas extraction. During the drilling of gas extraction holes, the drill rig's vibration signal changes significantly with the lithology of the drilling medium. These signals can be used to analyze the coal and rock properties in the drilling area, determine the construction status, and estimate the gas content, which is of great significance for ensuring safe production in coal mines. However, the complex construction environment and geological conditions in underground coal mines make the drill rig's vibration signal extremely susceptible to noise interference, seriously affecting the accuracy of subsequent signal analysis, feature extraction, and coal and rock property identification. Therefore, conducting research on noise reduction of drilling rig vibration signals is not only of great theoretical value, but also of practical significance for improving the safety and efficiency of coal mine production.

[0003] Variational Mode Decomposition (VMD), a non-recursive adaptive signal decomposition algorithm, has been widely used in the field of drilling rig vibration signal denoising due to its robustness and mathematical support. However, the denoising effect of VMD is highly dependent on parameter settings, particularly the selection of the mode number K and the penalty factor α. Existing VMD parameter optimization methods are mostly single-objective optimization methods, which optimize the VMD decomposition effect at a single scale and fail to fully reflect the multidimensional characteristics of complex signals. For complex signals such as drilling rig vibration signals, which contain both impact and non-impact components, traditional single-objective optimization methods often fail to balance the characteristics of both components, resulting in poor signal decomposition and, in turn, poor denoising performance. Furthermore, the intrinsic mode function (IMF) components obtained from VMD decomposition contain different original signal components, and the selection of effective modes directly determines the VMD denoising effect. Due to the nonlinear and non-stationary characteristics of drilling rig vibration signals, traditional single-objective metrics cannot fully describe the signal characteristics, resulting in poor IMF component screening and further limiting the improvement of denoising performance.

[0004] In actual drilling operations, drilling rig operations include various working conditions, such as drilling, retraction, and drill rod engagement, each of which generates different vibration signals. To ensure the reliability and effectiveness of subsequent denoising and coal and rock property identification, it is necessary to accurately identify the specific working conditions of the drilling rig and extract the drilling signal before noise reduction. However, existing technologies still have shortcomings in working condition identification and signal extraction, making it difficult to meet the high-precision requirements under complex working conditions. Therefore, how to combine advanced signal processing technology to construct a noise reduction method that can adaptively optimize parameters, effectively screen modes, and accurately extract drilling signals has become a technical challenge that needs to be solved urgently.

[0005] To address the above problems, the present invention proposes an adaptive denoising method for vibration signals of gas extraction drilling rigs based on multi-scale feature fusion. By introducing a target detection algorithm to achieve accurate segmentation of drilling signals, a multi-scale objective function is constructed to optimize VMD parameters, and a multi-feature fusion strategy is used to screen effective modes, thereby significantly improving the denoising effect of drilling rig vibration signals and the coal and rock identification accuracy. Summary of the Invention

[0006] In an embodiment of the present invention, an adaptive noise reduction method for vibration signals of gas extraction drilling rigs based on multi-scale feature fusion is provided to solve the problems in the prior art that drilling rig vibration signals are susceptible to noise interference, the accuracy of coal and rock property recognition is insufficient, and the noise reduction effect is poor due to the single parameter optimization.

[0007] An embodiment of the present invention provides a method for adaptively reducing noise of vibration signals of a gas extraction drilling rig based on multi-scale feature fusion, the method comprising the following steps:

[0008] S1. During the gas extraction drilling process, the vibration signal of the drilling rig is collected to obtain the vibration signal sequence at the current moment and its historical moments;

[0009] S2. Converting the vibration signal sequence into a two-dimensional waveform image, and using a pre-trained target detection algorithm to identify the drilling condition area in the image, and extracting the drilling vibration signal within the corresponding time range;

[0010] S3. Construct a multi-scale objective function that includes energy distribution and transient impact characteristics, establish a multi-objective optimization model for variational mode decomposition (VMD) parameters, and solve the model using the non-dominated sorting genetic algorithm II (NSGA-II) to obtain the optimal combination of the mode number K and the penalty factor α.

[0011] S4. Decomposing the drilling vibration signal using the optimized VMD parameters to obtain a set of intrinsic mode function (IMF) components;

[0012] S5. A multi-feature fusion strategy based on energy entropy and kurtosis is used to comprehensively evaluate each IMF component, select the dominant IMF component, reconstruct the signal, and output the denoised drilling vibration signal.

[0013] Preferably, the step S2 converts the vibration signal sequence into a two-dimensional waveform image, and uses a pre-trained target detection algorithm to identify the drilling condition area in the image, and extracts the drilling vibration signal within the corresponding time range, including:

[0014] S2.1. For the collected one-dimensional vibration signal, plot a time domain waveform graph with time as the horizontal axis and amplitude as the vertical axis to generate a two-dimensional waveform image;

[0015] S2.2. Use the pre-trained Yolov8n model to analyze the waveform image and identify the rectangular bounding box corresponding to the drilling condition. Its pixel coordinates are expressed as (x min ,y min ,x max ,y max ), where x min 、y min is the horizontal boundary, x max ,y max is the vertical boundary;

[0016] S2.3. Calculate the mapping relationship between time and pixels based on the time range [T1, T2] of the entire image and the total number of pixels N, and determine the actual time range [t min ,t max ], the mapping relationship between time and pixels is as follows:

[0017]

[0018] Where, t min ,t max They represent the start time and end time of the drilling action of the drilling rig respectively; N is the total number of pixels in the image;

[0019] S2.4. Intercept the time period [t min ,t max ] as the drilling vibration signal.

[0020] Preferably, the multi-scale objective function including energy distribution and transient impact characteristics is constructed in S3, including:

[0021] S3.1. Define the first objective function f1 as the sum of the energy entropy of all IMF components, which is used to describe the concentration of the signal energy distribution in the frequency domain. The calculation formula is as follows:

[0022]

[0023] Where H k is the energy entropy of the kth IMF component, E k is the energy of the kth IMF component, P k is the proportion of its total energy;

[0024] S3.2. Define the second objective function f2 as the sum of the kurtosis of all IMF components, which is used to describe the significance of the transient impact component in the signal. The calculation formula is as follows:

[0025]

[0026] Where Q k is the kurtosis of the kth IMF component, μ k and σ k are the mean and standard deviation of the kth IMF component respectively, and E is the mathematical expectation;

[0027] S3.3. The two objective functions together constitute a multi-scale objective function, which is used to reflect the overall characteristics of the signal.

[0028] Preferably, in S3, the model is solved by a non-dominated sorting genetic algorithm II (NSGA-II) to obtain an optimal combination of the modal number K and the penalty factor α, including:

[0029] S3.4. Initialize the population size, number of iterations, and crossover probability of the NSGA-II algorithm, and set the decision variables to the modal number K and the penalty factor α;

[0030] S3.5. In each iteration, calculate the objective function values f1 and f2 corresponding to each individual, and select the next generation population through non-dominated sorting and crowding distance;

[0031] S3.6. After multiple iterations, the Pareto optimal solution set is obtained;

[0032] S3.7. Use the TOPSIS method to comprehensively evaluate the solutions in the Pareto optimal solution set and select the optimal parameter combination (K, α).

[0033] Preferably, the multi-feature fusion strategy based on energy entropy and kurtosis in S5 performs a comprehensive evaluation on each IMF component, including:

[0034] S5.1. Calculate the energy entropy and kurtosis value of each IMF component as evaluation indicators;

[0035] S5.2. Use the TOPSIS method to score each IMF component. The specific steps include: constructing an evaluation matrix, normalizing the data, determining the positive ideal solution and the negative ideal solution, and calculating the distance between each IMF component and the positive ideal solution and the negative ideal solution;

[0036] S5.3. Based on the scoring results, select the IMF components with the highest scores as the dominant components.

[0037] Preferably, selecting the dominant IMF component and performing signal reconstruction in S5 includes:

[0038] S5.4. Superimpose the top M IMF components with the highest scores to obtain a reconstructed signal;

[0039] S5.5. Perform time domain and frequency domain analysis on the reconstructed signal to verify whether it retains the main features of the original signal and removes the noise components.

[0040] Preferably, the step of outputting the drilling vibration signal after noise reduction in S5 includes:

[0041] S5.6. Compare the reconstructed signal with the original signal and calculate the waveform similarity, power spectrum density overlap, and correlation coefficient;

[0042] S5.7. If the above indicators meet the preset thresholds, the noise reduction is determined to be successful and the final result is output; otherwise, the parameters are adjusted and optimized again.

[0043] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0044] The present invention combines the energy distribution of the signal with the transient impact characteristics through the design of a multi-scale objective function, thus solving the problem that traditional single-objective optimization methods are difficult to take into account the multi-dimensional characteristics of complex signals. The NSGA-II algorithm combined with the TOPSIS method is used to optimize the VMD parameters, which can effectively improve the decomposition effect and make the IMF component more accurately reflect the essential characteristics of the signal. In addition, the multi-feature fusion strategy based on energy entropy and kurtosis realizes the precise screening of IMF components, avoiding the limitations of single indicator evaluation. Experiments show that the method proposed in the present invention can significantly improve the waveform similarity, power spectrum density overlap and correlation coefficient when processing drilling rig vibration signals, while improving the accuracy of coal rock property identification, providing a reliable technical guarantee for coal mine safety production. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Attachment Figure 1 is the overall flow chart of the present invention,

[0046] Attachment Figure 2 Schematic diagram of vibration signal acquisition and waveform image generation;

[0047] Attachment Figure 3 A diagram showing the process of identifying the drilling area for the target detection algorithm;

[0048] Attachment Figure 4It is a structural diagram of the solution process of the multi-objective optimization model;

[0049] Attachment Figure 5 Flowchart for IMF component evaluation and signal reconstruction;

[0050] Attachment Figure 6 Schematic diagram of the drilling rig and measuring points in the embodiment;

[0051] Figure 7 is a waveform diagram of the drilling rig vibration signal collected at measuring points 1 and 2 in the embodiment;

[0052] Attachment Figure 8 The drilling signal recognition result based on Yolov8n in the embodiment;

[0053] Attachment Figure 9 This is a flow chart of optimizing (K, α) parameters of NSGA-Ⅱ in the embodiment;

[0054] Figure 10 is the Pareto frontier obtained by solving the VMD parameters using NSGA-II in the embodiment;

[0055] Attachment Figure 11 Graph showing the evaluation scores of the various IMF components in the embodiment;

[0056] Figure 12 is a comparison diagram of the drilling signal waveforms of the drilling rig before and after denoising in the embodiment;

[0057] Attachment Figure 13 3 is a comparison diagram of the drilling signal PSD of the drilling rig before and after denoising in the embodiment. DETAILED DESCRIPTION

[0058] The embodiment of the present invention provides a method for adaptive noise reduction of vibration signals of gas extraction drilling rig based on multi-scale feature fusion. Figure 1 To the attached Figure 13 The following is a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings and the component numbers involved therein.

[0059] In practical applications, it is first necessary to collect the vibration signal of the drilling rig during the gas extraction drilling process. Figure 2As shown, the vibration signal acquisition device is connected to the drilling rig via a sensor. The sensor is positioned close to the drill rod to ensure accurate capture of vibration signals generated during drilling operation. The sensor transmits the collected vibration signals to the data acquisition module, which records the current and historical vibration signal sequences. The data acquisition module uses a signal conditioning circuit to perform preliminary filtering on the raw signal to remove high-frequency interference and DC offset, ensuring data quality for subsequent processing. The signal conditioning circuit is connected to an analog-to-digital converter, which converts the analog signal into a digital signal and stores it in a storage unit. The storage unit communicates with the central processing unit via a data interface, which is responsible for subsequent signal processing and analysis.

[0060] Next, the collected one-dimensional vibration signal is converted into a two-dimensional waveform image, and the pre-trained target detection algorithm is used to identify the drilling condition area in the image. Figure 3 As shown, the central processing unit calls the image generation module, which plots the vibration signal data in the storage unit with time as the horizontal axis and amplitude as the vertical axis in the time domain waveform to generate a two-dimensional waveform image. The generated image is sent to the target detection module, which uses the pre-trained Yolov8n model to analyze the image. The Yolov8n model extracts image features through a convolutional neural network and identifies the rectangular bounding box corresponding to the drilling condition, whose pixel coordinates are expressed as (xmin, ymin, xmax, ymax). The target detection module sends the recognition result to the time mapping module, which calculates the mapping relationship between time and pixels based on the time range [T1, T2] of the entire image and the total number of pixels N, and determines the actual time range [t min ,t max Subsequently, the time mapping module sends an instruction to the storage unit through the data interface to intercept the time period [t min ,t max ] is used as the drilling vibration signal.

[0061] After completing the extraction of drilling vibration signals, a multi-scale objective function including energy distribution and transient impact characteristics is constructed, and a multi-objective optimization model of variational mode decomposition parameters is established. Figure 4 As shown, the central processing unit calls the objective function construction module, which defines the sum of the energy entropy of all IMF components of the first objective function f1, which is used to describe the concentration of the signal energy distribution in the frequency domain. The calculation formula of the first objective function f1 is as follows: Where H k is the energy entropy of the kth IMF component, E k is the energy of the kth IMF component, P kis the proportion of its total energy. At the same time, the objective function construction module 14 defines the second objective function f2 as the sum of the kurtosis of all IMF components, which is used to describe the significance of the transient impact component in the signal. The calculation formula of the second objective function f2 is as follows: Where Q k is the kurtosis of the kth IMF component, μ k and σ k are the mean and standard deviation of the kth IMF component, respectively, and E is the mathematical expectation. These two objective functions together constitute a multiscale objective function, which reflects the overall characteristics of the signal. The objective function construction module passes the multiscale objective function to the optimization module, which initializes the population size, number of iterations, and crossover probability of the non-dominated sorting genetic algorithm II (NSGA-II), setting the decision variables to the modal number K and the penalty factor α. During each iteration, the optimization module calculates the objective function values f1 and f2 corresponding to each individual and selects the next generation of the population based on the non-dominated sorting and crowding distance. After multiple iterations, a Pareto optimal solution set is obtained. The optimization module further calls the comprehensive evaluation module, which uses the TOPSIS method to comprehensively evaluate the solutions in the Pareto optimal solution set and select the optimal parameter combination (K, α).

[0062] After obtaining the optimal VMD parameters, the optimized parameters are used to decompose the drilling vibration signal to obtain a set of intrinsic mode function (IMF) components. Figure 5 As shown, the central processing unit (CPU) invokes the signal decomposition module, which loads the parameter combination (K, α) output by the optimization module and performs variational modal decomposition on the drilling vibration signal, generating multiple IMF components and storing them in the IMF component storage unit. Subsequently, the CPU invokes the feature evaluation module, which comprehensively evaluates each IMF component based on a multi-feature fusion strategy of energy entropy and kurtosis. Specifically, the feature evaluation module calculates the energy entropy and kurtosis values of each IMF component as evaluation indicators. Next, the feature evaluation module uses the TOPSIS method to score each IMF component. The scoring process includes constructing an evaluation matrix, normalizing the data, determining the positive and negative ideal solutions, and calculating the distance between each IMF component and the positive and negative ideal solutions. Based on the scoring results, the feature evaluation module selects the IMF components with the highest scores as the dominant components and sends this dominant component information to the signal reconstruction module.

[0063] The signal reconstruction module receives the dominant component information output by the feature evaluation module and superimposes the top M IMF components with the highest scores to obtain a reconstructed signal. The reconstructed signal is sent to the signal verification module, which performs time and frequency domain analysis on the reconstructed signal to verify whether it retains the main features of the original signal and removes the noise components. The signal verification module determines whether the noise reduction effect meets the requirements by calculating indicators such as waveform similarity, power spectral density overlap, and correlation coefficient. If the above indicators meet the preset threshold, the signal verification module determines that the noise reduction is successful and outputs the final result to the result storage unit. Otherwise, the signal verification module sends a feedback signal to the optimization module, triggering the parameter adjustment process and re-optimization.

[0064] Throughout the implementation process, each module communicates with each other via data interfaces, ensuring efficient data flow. For example, communication between the object detection module and the time mapping module relies on the data interface, through which the time mapping module reads raw vibration signal data from the storage unit. Similarly, parameter transfer between the optimization module and the signal decomposition module is also achieved through the data interface. Furthermore, the central processing unit, serving as the core control unit, coordinates the working sequence of each module to ensure the orderly execution of the entire process. For example, once the objective function construction module completes the construction of the multi-scale objective function, the central processing unit immediately activates the optimization module to perform parameter optimization.

[0065] In practical application scenarios, the method of the present invention can be applied to the drilling rig vibration signal processing during the coal mine gas extraction drilling construction process. For example, at a coal mine construction site, the drilling rig body performs gas extraction operations through the drill rod, and the sensor collects the vibration signal of the drilling rig in real time. The data acquisition module records the vibration signal sequence and stores it in the storage unit. Subsequently, the central processing unit calls the image generation module and the target detection module to identify the time range corresponding to the drilling condition [t min ,t max ]. The optimization module obtains the optimal VMD parameter combination (K, α) by solving the multi-scale objective function, and the signal decomposition module decomposes the drilling vibration signal. The feature evaluation module selects the dominant IMF components based on the multi-feature fusion strategy of energy entropy and kurtosis. The signal reconstruction module superimposes the dominant components to generate the drilling vibration signal after noise reduction. The signal verification module verifies whether the noise reduction effect meets the requirements by calculating indicators such as waveform similarity, power spectrum density overlap, and correlation coefficient. If the requirements are met, the result storage unit saves the final result for subsequent coal and rock property identification and analysis.

[0066] It can be seen from the above embodiments that the present invention combines the energy distribution of the signal with the transient impact characteristics through the design of a multi-scale objective function, thereby solving the problem that traditional single-objective optimization methods are difficult to take into account the multi-dimensional characteristics of complex signals. The NSGA-II algorithm combined with the TOPSIS method is used to optimize the VMD parameters, which can effectively improve the decomposition effect and make the IMF component more accurately reflect the essential characteristics of the signal. In addition, the multi-feature fusion strategy based on energy entropy and kurtosis realizes the precise screening of IMF components, avoiding the limitations of single indicator evaluation. Experiments show that the method proposed in the present invention can significantly improve the waveform similarity, power spectrum density overlap and correlation coefficient when processing drilling rig vibration signals, while improving the accuracy of coal rock property identification, providing a reliable technical guarantee for coal mine safety production.

[0067] In order to better enable relevant personnel in this technical field to fully understand and implement the present invention, the specific implementation principle of the present invention is further supplemented below with reference to a specific application scenario.

[0068] An adaptive noise reduction method for vibration signals of gas extraction drilling rigs based on multi-scale feature fusion is proposed. Figure 1 As shown, the specific steps of this method are:

[0069] S1. During the gas extraction drilling process, the vibration signal of the drilling rig is collected to obtain the vibration signal sequence at the current moment and its historical moments;

[0070] S2. Converting the vibration signal sequence into a two-dimensional waveform image, and using a pre-trained target detection algorithm to identify the drilling condition area in the image, and extracting the drilling vibration signal within the corresponding time range;

[0071] The vibration forms of drill pipe are mainly divided into three types: axial vibration, lateral vibration and torsional vibration. Here we focus on the axial and lateral vibration of the drill pipe, and the vibration measurement points are arranged as follows: Figure 6 As shown, measuring point 1 and measuring point 2 are used to collect lateral and axial vibration signals respectively.

[0072] The waveforms of the drilling rig vibration signals collected at measuring points 1 and 2 are shown in Figures 7(a) and 7(b), respectively. During the gas extraction hole drilling process, the waveforms of the vibration signals collected at measuring points 1 and 2 show essentially the same variation, but there are significant differences in the signal waveforms under different drilling rig operating conditions. Taking the signal collected at measuring point 1 as an example, when the drill rig begins to engage the drill pipe, a high-amplitude vibration acceleration signal is generated, with an amplitude approaching 200 m / s². However, when the drill pipe is withdrawn, the vibration amplitude decreases significantly, with a peak amplitude of only 3.2 m / s². During the drilling operation, the drill rig vibration signal exhibits continuous spikes due to the interaction between the drill bit and the rock formation.

[0073] Multiple experiments were conducted to collect drilling rig vibration signals and construct a dataset of gas extraction borehole drilling rig vibration signal waveform images. The dataset was divided into training, validation, and test sets using a simple random partitioning method. The training, validation, and test sets had a ratio of 0.8:0.08:0.12.

[0074] The target detection algorithm Yolov8n is trained using the above dataset, and the optimal weight after training is used to detect and predict the drilling signal. The recognition results are as follows: Figure 8 As shown in the figure, under the condition that the detection duration is known, the time period of the drilling condition can be determined based on the detection frame coordinates, thus realizing the segmentation of the drilling signal. Figure 8 In the image, the detection duration of the signal is 52 seconds. Based on the pixel ratio, it can be determined that the drilling signal is located between 36.4 seconds and 42.2 seconds and between 46.3 seconds and 52.0 seconds.

[0075] S3. Construct a multi-scale objective function that includes energy distribution and transient impact characteristics, establish a multi-objective optimization model for variational mode decomposition (VMD) parameters, and solve the model using the non-dominated sorting genetic algorithm II (NSGA-II) to obtain the optimal combination of the mode number K and the penalty factor α.

[0076] The multi-scale objective function is as follows:

[0077]

[0078] Where f1 is the sum of the energy entropy of the IMF components, and f2 is the sum of the kurtosis of the IMF components. k is the kurtosis of the kth IMF component uk, μ k and σ k Represents the kth IMF component u k The mean and variance of H, E is the mathematical expectation. k is the energy entropy of the kth IMF component, E k 、P k are the kth IMF components u k The energy and total energy ratio is calculated as follows:

[0079]

[0080] The VMD parameter multi-objective optimization model is as follows:

[0081]

[0082] To demonstrate the effectiveness of the proposed denoising algorithm, a Gaussian white noise with SNR = 0dB is added to the sampling frequency band to obtain the noisy signal Sn1(t). Sn1(t) is used as input and NSGA-II is used to find the optimal VMD parameters (K, α) based on the MATLAB platform, as shown in Figure 10. The Pareto frontier is as follows: Figure 11 As shown in the figure, the TOPSIS method was used to determine the optimal (K, η) parameter combination to be (22, 3910).

[0083] S4. Decomposing the drilling vibration signal using the optimized VMD parameters to obtain a set of intrinsic mode function (IMF) components;

[0084] S5. A multi-feature fusion strategy based on energy entropy and kurtosis is used to comprehensively evaluate each IMF component, select the dominant IMF component, reconstruct the signal, and output the denoised drilling vibration signal.

[0085] The energy entropy and kurtosis values of each IMF component were calculated, and the TOPSIS method was used to evaluate the IMF components. The IMF evaluation scores are shown in Figure 12. Using a threshold of 0.5, IMF components were screened. Specifically, IMF components with an evaluation score exceeding 0.5 were considered signal-dominant modes and retained; those with an evaluation score exceeding 0.5 were considered noise-dominant modes and removed. Based on the evaluation scores, IMF2, IMF4, IMF5, and IMF7 were selected as valid modes for signal reconstruction.

[0086] The waveforms of the original signal S1(t), the noisy signal Sn1(t) and the denoised signal Sd1(t) are shown in Figure 12(a), Figure 12(b) and Figure 12(c) respectively, and their corresponding PSDs are as follows: Figure 13 shown.

[0087] In order to demonstrate the advantages of the denoising method based on multi-scale feature fusion proposed in the present invention in processing complex signals, it is compared with the currently commonly used VMD parameter single-objective optimization method and effective mode screening method. The objective function is constructed based on energy entropy and kurtosis, and the VMD parameters are optimized using single-objective optimization algorithms - genetic algorithm (GA), gray wolf optimization algorithm (GWO), and whale optimization algorithm (WOA). The noisy signal Sn1(t) is decomposed using the VMD after parameter optimization to obtain K IMF components. Effective mode screening is performed based on correlation coefficient, energy ratio, and variance contribution rate, respectively. Among them, the threshold of the correlation coefficient method is set to 0.5, and the energy ratio threshold is set to 50%.

[0088] To quantitatively evaluate the denoising effects of the above denoising methods, the correlation coefficient R and coal-rock identification accuracy are introduced to evaluate the denoising effects. The comparison of the denoising effects of different methods is shown in Table 1.

[0089] Table 1 Comparison of denoising effects of different methods

[0090]

[0091] Experiments have shown that the adaptive denoising method for drilling rig vibration signals proposed in this paper achieves excellent results. In comparative experiments, the proposed method outperforms existing common denoising algorithms in terms of correlation coefficient and recognition accuracy. The proposed method effectively eliminates noise interference while preserving the waveform characteristics of the drilling signal. The denoised signal clearly demonstrates the actual changes in the drilling signal.

[0092] It can be seen from the above embodiments that the present invention combines the energy distribution of the signal with the transient impact characteristics through the design of a multi-scale objective function, thereby solving the problem that traditional single-objective optimization methods are difficult to take into account the multi-dimensional characteristics of complex signals. The NSGA-II algorithm combined with the TOPSIS method is used to optimize the VMD parameters, which can effectively improve the decomposition effect and make the IMF component more accurately reflect the essential characteristics of the signal. In addition, the multi-feature fusion strategy based on energy entropy and kurtosis realizes the precise screening of IMF components, avoiding the limitations of single indicator evaluation. Experiments show that the method proposed in the present invention can significantly improve the waveform similarity, power spectrum density overlap and correlation coefficient when processing drilling rig vibration signals, while improving the accuracy of coal rock property identification, providing a reliable technical guarantee for coal mine safety production.

Claims

1. An adaptive noise reduction method for vibration signals of gas extraction drilling rigs based on multi-scale feature fusion, characterized in that: The method comprises the following steps: S1. During the gas extraction drilling process, the vibration signal of the drilling rig is collected to obtain the vibration signal sequence at the current moment and its historical moments; S2. Converting the vibration signal sequence into a two-dimensional waveform image, and using a pre-trained target detection algorithm to identify the drilling condition area in the image, and extracting the drilling vibration signal within the corresponding time range; S3. Construct a multi-scale objective function that includes energy distribution and transient impact characteristics, establish a multi-objective optimization model for variational mode decomposition (VMD) parameters, and solve the model using the non-dominated sorting genetic algorithm II (NSGA-II) to obtain the optimal combination of the mode number K and the penalty factor α. S4. Decomposing the drilling vibration signal using the optimized VMD parameters to obtain a set of intrinsic mode function (IMF) components; S5. A multi-feature fusion strategy based on energy entropy and kurtosis is used to comprehensively evaluate each IMF component, select the dominant IMF component, reconstruct the signal, and output the denoised drilling vibration signal.

2. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 1 is characterized in that: In S2, the vibration signal sequence is converted into a two-dimensional waveform image, and the drilling condition area in the image is identified using a pre-trained target detection algorithm to extract the drilling vibration signal within the corresponding time range, including: S2.

1. For the collected one-dimensional vibration signal, plot a time domain waveform graph with time as the horizontal axis and amplitude as the vertical axis to generate a two-dimensional waveform image; S2.

2. Use the pre-trained Yolov8n model to analyze the waveform image and identify the rectangular bounding box corresponding to the drilling condition. Its pixel coordinates are expressed as (x min ,y min ,x max ,y max ), where x min 、y min is the horizontal boundary, x max ,y max is the vertical boundary; S2.

3. According to the time range [T1, T2] of the entire image and the total number of pixels N, calculate the mapping relationship between time and pixels and determine the actual time range [t min ,t max ], the mapping relationship between time and pixels is as follows: Where, t min ,t max They represent the start time and end time of the drilling action of the drilling rig respectively; N is the total number of pixels in the image; S2.

4. Intercept the time period [t min ,t max ] as the drilling vibration signal.

3. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 1 is characterized in that: In S3, a multi-scale objective function including energy distribution and transient impact characteristics is constructed, including: S3.

1. Define the first objective function f1 as the sum of the energy entropy of all IMF components, which is used to describe the concentration of the signal energy distribution in the frequency domain. The calculation formula is as follows: Where H k is the energy entropy of the kth IMF component, E k is the energy of the kth IMF component, P k is the proportion of its total energy; S3.

2. Define the second objective function f2 as the sum of the kurtosis of all IMF components, which is used to describe the significance of the transient impact component in the signal. The calculation formula is as follows: Where Q k is the kurtosis of the kth IMF component, μ k and σ k are the mean and standard deviation of the kth IMF component respectively, and E is the mathematical expectation; S3.

3. The two objective functions together constitute a multi-scale objective function, which is used to reflect the overall characteristics of the signal.

4. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 1 is characterized in that: In S3, the model is solved by a non-dominated sorting genetic algorithm II (NSGA-II) to obtain the optimal combination of the modal number K and the penalty factor α, including: S3.

4. Initialize the population size, number of iterations, and crossover probability of the NSGA-II algorithm, and set the decision variables to the modal number K and the penalty factor α; S3.

5. In each iteration, calculate the objective function values f1 and f2 corresponding to each individual, and select the next generation population through non-dominated sorting and crowding distance; S3.

6. After multiple iterations, the Pareto optimal solution set is obtained; S3.

7. Use the TOPSIS method to comprehensively evaluate the solutions in the Pareto optimal solution set and select the optimal parameter combination (K, α).

5. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 1 is characterized in that: In S5, a multi-feature fusion strategy based on energy entropy and kurtosis is used to comprehensively evaluate each IMF component, including: S5.

1. Calculate the energy entropy and kurtosis value of each IMF component as evaluation indicators; S5.

2. Use the TOPSIS method to score each IMF component. The specific steps include: constructing an evaluation matrix, normalizing the data, determining the positive ideal solution and the negative ideal solution, and calculating the distance between each IMF component and the positive ideal solution and the negative ideal solution; S5.

3. Based on the scoring results, select the IMF components with the highest scores as the dominant components.

6. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 1, characterized in that: The S5 selects the dominant IMF component and performs signal reconstruction, including: S5.

4. Superimpose the top M IMF components with the highest scores to obtain a reconstructed signal; S5.

5. Perform time domain and frequency domain analysis on the reconstructed signal to verify whether it retains the main features of the original signal and removes the noise components.

7. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 1 is characterized in that: Outputting the drilling vibration signal after noise reduction in S5 includes: S5.

6. Compare the reconstructed signal with the original signal and calculate the waveform similarity, power spectrum density overlap, and correlation coefficient; S5.

7. If the above indicators meet the preset thresholds, the noise reduction is determined to be successful and the final result is output; otherwise, the parameters are adjusted and optimized again.

8. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 2 is characterized in that: The pre-trained Yolov8n model extracts image features through a convolutional neural network and identifies the rectangular bounding box corresponding to the drilling condition.

9. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 4 is characterized in that: The TOPSIS method determines the optimal parameter combination (K, α) by calculating the distance between each solution and the positive ideal solution and the negative ideal solution.

10. The method for adaptive noise reduction of vibration signals of gas extraction drilling rigs based on multi-scale feature fusion according to claim 5, characterized in that: The number M of the dominant components is a preset fixed value.

Images