Multi-channel signal decomposition parameter adaptive optimization noise reduction method based on enhanced dung beetle optimization algorithm

Abstract

The invention provides a multichannel signal decomposition parameter adaptive optimization noise reduction method based on an enhanced dung beetle optimization algorithm. The method comprises the following steps: S1, designing the enhanced dung beetle optimization algorithm to optimize a parameter combination; s2, the algorithm is initialized; s3, substituting the weight coefficient and the initial value of the smoothness control parameter into a JMCMD method, and decomposing the original noisy signal; s4, constructing a time-frequency energy distribution matrix according to the decomposed signal components, and evaluating the energy concentration degree by using a target function; s5, performing iteration according to the fitness value; s6, when the maximum number of iterations is reached, determining an optimal parameter combination; and S7, substituting the optimal parameter into a JMCMD method and processing to obtain a signal with a high signal-to-noise ratio. The method has the beneficial effects that the defects of manual parameter adjustment are avoided, and the usability and reliability of the method are improved; chaos initialization and a nonlinear R factor can improve the capability of jumping out of local optimum and have higher optimization precision; and good signal separation and noise reduction effects are achieved.

Description

Technical Field

[0001] This invention belongs to the field of signal processing and fault diagnosis technology, specifically relating to a method for efficient noise reduction of multi-channel non-stationary vibration signals such as those from hydroelectric generator sets by automatically finding optimal parameters using intelligent optimization algorithms. Background Technology

[0002] Hydropower generating units are core equipment in the national energy strategy, and the vibration signals they generate during operation are key indicators for assessing the unit's health status. However, due to variable operating conditions, environmental noise interference, and coupled vibrations from multiple components, the collected signals often exhibit characteristics of multi-channel operation, strong non-stationarity, and low signal-to-noise ratio, posing a significant challenge to the accurate extraction of fault features.

[0003] Huang Jie's paper, "Joint extraction-based multivariate chirp mode decomposition and its application to fault diagnosis of rotating machinery under variable speed conditions," proposes a joint extraction-based multivariate chirp mode decomposition (JMCMD) method, which can effectively handle such signals. However, since JMCMD inherently uses a joint decomposition framework, this type of method typically includes some key parameters that decisively influence the decomposition effect, such as weighting coefficients and smoothness control parameters. In existing technologies, these parameters mostly rely on engineering experience or manual trial and error to set, which has the following drawbacks: (1) Highly subjective: The selection of parameters depends heavily on the user's professional knowledge and lacks universality.

[0004] (2) Inefficient: Repeated trial and error is time-consuming and labor-intensive, and it is difficult to guarantee that the parameter combination is the global optimum.

[0005] (3) Limited effect: Suboptimal parameter combinations can lead to modal aliasing, incomplete decomposition or distortion of useful signals, ultimately affecting the noise reduction effect and the accuracy of subsequent fault diagnosis.

[0006] Although some studies have attempted to use standard intelligent optimization algorithms (such as particle swarm optimization and genetic algorithms) for parameter optimization, standard algorithms (including the standard dung beetle optimization algorithm) also have problems such as being prone to getting trapped in local optima and an imbalance between convergence speed and optimization accuracy. In particular, their performance still needs to be improved when dealing with complex, high-dimensional optimization problems.

[0007] Therefore, there is an urgent need for a new method that can automatically, quickly, and accurately find the optimal parameter combination for signal decomposition algorithms and thereby achieve excellent noise reduction results. Summary of the Invention

[0008] The main objective of this invention is to provide an adaptive optimization noise reduction method for channel signal decomposition parameters based on an enhanced dung beetle optimization algorithm, thereby solving the problems mentioned in the background art.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a multi-channel signal decomposition parameter adaptive optimization noise reduction method based on the enhanced dung beetle optimization algorithm, comprising the following steps: S1. Design an enhanced dung beetle optimization algorithm to optimize the weight coefficients in the Jointly Extracted Multivariate Frequency Modulation Mode Decomposition (JMCMD) method. and smoothness control parameters Optimize; S2. Initialize the enhanced dung beetle optimization algorithm to generate an initial population. The position of each individual in the population is the initial value of the parameter to be optimized. Initialization includes setting the population size, maximum number of iterations, and search space for the parameters to be optimized in the enhanced dung beetle optimization algorithm; S3. Substitute the initial values ​​of the weighting coefficients and smoothness control parameters into the JMCMD method to decompose the original noisy signal; S4. Construct a time-frequency energy distribution matrix based on the decomposed signal components, and use an objective function to evaluate the energy concentration of the energy distribution matrix; The result of the objective function is expressed in the form of fitness values; S5. Iterate based on fitness values. In each iteration, use update rules to update the position of the entire population. S6. When the maximum number of iterations is reached, the individual with the lowest fitness value in the population is the globally optimal combination of parameters to be optimized. S7. Substitute the found optimal parameters into the JMCMD method to decompose the original noisy signal. By analyzing the permutation entropy or energy of each mode, identify and remove the noise modes, and superimpose the remaining effective signal modes to obtain a high signal-to-noise ratio signal after noise reduction.

[0010] Furthermore, the process of decomposing the signal using the JMCMD method is as follows: S201. Modeling of multivariable linear frequency modulation modes: ; in, The signal to be decomposed, Number of signal channels , , They represent the first The instantaneous amplitude, instantaneous frequency, and initial phase of each channel; S202. Treating the multi-channel signal as a whole to simplify the model, the above equation can be rewritten as: ; ; ; in, It is white noise and follows a normal distribution. , For the first The instantaneous frequency and demodulation frequency of the first mode. and For the first The demodulated signal components of the first mode, and For the demodulated signal, the first Channel 1 The demodulated signal components of the first mode, The total number of modes, Indicates the first Class modality; S203. Obtain the initial value of the instantaneous frequency of the signal; S204. Design constraints to extract signal components from the signal to be decomposed in the form that best meets the constraints, thereby achieving signal decomposition; assuming the signal sampling time is: ; The constraints are as follows: ; ; ; ; ; ; in, This represents the total number of samples. For the first The signal components of the first mode of the channel, For demodulated signals, It is a second-order difference operator. It is a norm 2; S205. The instantaneous frequency is updated using arctangent demodulation technology. The frequency update formula is: ; in, For the first The first update Channel 1 The instantaneous frequency of the modal signal components, It is the identity matrix. For the first The first update Channel 1 The instantaneous frequency increment of the modal signal component, , For the first The demodulated signal at the next update Channel 1 Demodulated signal components of the first mode; To simplify the multi-channel input signal model, we assume that all channels have a common frequency component, then: ; in, No. The first update Modal multi-channel common instantaneous frequency components, For the first The first update Channel 1 Instantaneous amplitude of the mode; S206. Update the decomposed signal: ; in, For the first The demodulated signal of the next update. For the first The updated matrix ; S207. Repeat steps S205 to S206 until the stopping condition is met.

[0011] Furthermore, the detailed process of evaluating the objective function is as follows: S301. Perform discrete short-time Fourier transform on the signal components: ; in, For the first The first sampling Channel 1 Modal signal components, The total number of samples, For the first Channel 1 Short-time Fourier transform of modal signal components, For window functions, For time frame indexing, Frequency point index; S302. Calculate the amplitude energy spectrum of the signal components as follows: ; in, For the first Channel 1 Time-frequency energy spectrum of modal signal components, for The complex conjugate; S303. By linearly superimposing the time-frequency energy spectra of all components, the time-frequency energy distribution matrix is ​​obtained as follows: ; in, The time-frequency energy distribution matrix is... The total number of channels. The total number of modes; S304. Arrange the elements in the time-frequency energy distribution matrix in ascending order and convert them into a one-dimensional vector: ; in, This represents the total number of elements in the time-frequency energy distribution matrix; S305. Calculate the cumulative energy and total energy of the vector: ; in, For the front The cumulative energy of each element, It is the total energy; At the same time, regulations =0; S306. Calculate the Gini coefficient based on the cumulative energy and the total energy. : ; When the total energy is 0, the Gini coefficient is defined as 1; S307. Use the Gini coefficient to represent the objective function: ; in, This is the fitness value.

[0012] Furthermore, the enhanced dung beetle optimization algorithm initializes the population using an improved Circle chaotic map, as shown in the following formula: ; in, For the first Chaotic values ​​generated in the next iteration The current chaos value, Indicates key control parameters. Indicates control parameters, The function represents the remainder operation. These represent the control parameters of the improved part.

[0013] Furthermore, the initial population includes proportionally divided individuals into rolling individuals, brooding individuals, offspring individuals, and thieving individuals; For individuals in the initial population Its current position is taken as the individual's best historical position. For all individuals, find the individual with the minimum fitness corresponding to the historical best position, and take that individual's historical best position as the global best position. .

[0014] Furthermore, the iterative process is as follows: S601. Update the positions of the rolling individual, the brooding individual, the offspring individual, and the stolen individual; For brooding individuals, the location update process is as follows: ; ; in, For the number of iterations, For the first The individual The position of the next iteration. , It is a random number. , These are the lower and upper boundaries of the egg cell's boundary. As the impact factor, , These are the lower and upper bounds of the parameter search space; For child individuals, the position update process is as follows: ; ; in, It is a random number that follows a normal distribution. It is a random number with a value in the range [0,1]. , These are the lower and upper bounds of the sub-individual's position, respectively; For the individual who commits the theft, the location update process is as follows: ; in, As a constant, It is a random value that follows a normal distribution; S602. Correct the updated position of each individual to ensure that each individual is within the boundary range. The correction process is as follows: when season ; when season ; S603. Using the corrected position as the weight coefficient and smoothness control parameter, execute steps S3~S4 to calculate the fitness value of each individual. S604. Update the historical best position and global best position of the individual based on the fitness value. The update process is as follows: For individuals Compare its first Fitness value of the next iteration The size of its fitness value relative to its historical best position, when season ; Find the individual with the smallest fitness value corresponding to the historical best position. Compare the individual's fitness value with the fitness value of the global optimal position. ,make .

[0015] Furthermore, the rolling ball individual calculates its updated position using both the variable spiral greedy selection method and the direct optimization movement method, and takes the position with the lower fitness value as the position update value.

[0016] Furthermore, the core of the variable spiral greedy selection method is a selection mechanism that varies with the number of iterations. Dynamically changing spiral shape parameters : ; in, This represents the maximum number of iterations. The formula for updating an individual's location is: ; ; in, In the variable spiral greedy selection method, the first The individual The individual position in the next iteration It is a random number within the interval [−1, 1].

[0017] Furthermore, the direct optimization movement method updates the individual position as follows: ; in, Direct Optimization Move Method The individual The individual position in the next iteration and These are the step size control coefficients, which are preset normal values.

[0018] Furthermore, the influence factor The expression is: ; in, This represents the maximum number of iterations.

[0019] Beneficial effects: 1. It completely solves the reliance on human experience in complex signal processing methods such as JMCMD, and realizes adaptive and optimal setting of key parameters, greatly improving the ease of use and reliability of the method.

[0020] 2. Chaotic initialization and nonlinear R-factor enhance the algorithm's ability to escape local optima and improve optimization accuracy, enabling JMCMD to find better parameter combinations than the standard algorithm.

[0021] 3. Since JMCMD operates under optimal parameters, the accuracy and thoroughness of its mode decomposition are guaranteed, and it can more effectively separate useful signal features from noise. The signal-to-noise ratio of the signal processed by this method is higher and the distortion is smaller.

[0022] 4. The optimal choice between the variable spiral greedy algorithm and the direct optimization method is made, which further improves the optimization accuracy of the dung beetle optimization algorithm. Attached Figure Description

[0023] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart illustrating the overall technical process of this invention; Figure 2 This is a comparison of the average convergence curves of the enhanced dung beetle optimization algorithm of this invention and the standard optimization algorithm on the simulation signal; Figure 3 These are the time-domain waveforms of each channel in the actual data and the corresponding FFT spectrum. Figure 4 It is the modal frequency domain diagram after the actual signal is decomposed. Detailed Implementation

[0024] Example 1 Reference Figure 1 The overall technical process of this paper, namely, an adaptive optimization denoising method for multi-channel signal decomposition parameters based on an enhanced dung beetle optimization algorithm, includes the following steps: S1. Design an Enhanced Dung Beetle Optimization (EDBO) algorithm to optimize the weight coefficients in the JMCMD method. and smoothness control parameters Optimize.

[0025] S2. Initialize the enhanced dung beetle optimization algorithm to generate an initial population. The position of each individual in the population is the initial value of the parameter to be optimized. Initialization includes setting the population size, maximum number of iterations, and search space for the parameters to be optimized in the enhanced dung beetle optimization algorithm; To improve the ergodicity and diversity of the initial population and avoid the algorithm from prematurely getting trapped in local optima due to uneven random initialization, this invention uses an improved Circle chaotic map to replace the random initialization method in the standard Dung Beetle Optimization (DBO) algorithm. Utilizing the ergodicity and uniformity of the chaotic sequence, the initial population can more broadly cover the entire solution space, effectively improving population diversity and laying the foundation for global optimization. The improved Circle chaotic map is used to initialize the population, and its formula is as follows: ; in, For the first Chaotic values ​​generated in the next iteration The current chaos value, Indicates key control parameters, which determine This item's "amplitude" This represents a control parameter, which can be viewed as an offset term or a phase factor. The function represents the remainder operation. The control parameters representing the improvements determine the strength of the newly added nonlinear terms; The chaos value generator calculates a new temporary value through a complex transformation involving linear growth, primary nonlinear oscillation, and secondary nonlinear perturbation. Finally, by taking the fractional part, this temporary value is "pushed back" to the standard interval to obtain the next chaos value. This process is repeated continuously, which produces a chaotic sequence that never repeats, appears random but is generated by deterministic rules, and this chaotic sequence is used as the initial position of the population. The initial population was divided proportionally into rolling individuals, brooding individuals, offspring individuals, and stealing individuals; For individuals in the initial population Its current position is taken as the individual's best historical position. For all individuals, find the individual with the minimum fitness corresponding to the historical best position, and take that individual's historical best position as the global best position. .

[0026] S3. Substitute the initial values ​​of the weighting coefficients and smoothness control parameters into the JMCMD method to decompose the original noisy signal. The process is as follows: A1. Modeling of multivariable linear frequency modulation modes: ; in, The signal to be decomposed, Number of signal channels , , They represent the first The instantaneous amplitude, instantaneous frequency, and initial phase of each channel; A2. By treating the multi-channel signal as a whole to simplify the model, the above equation can be rewritten as: ; ; ; in, It is white noise and follows a normal distribution. , For the first The instantaneous frequency and demodulation frequency of the first mode. and For the first The demodulated signal components of the first mode, and For the demodulated signal, the first Channel 1 The demodulated signal components of the first mode, The total number of modes, Indicates the first Class modality; A3. Use parametric unmodulation methods to obtain the initial value of the instantaneous frequency of the signal; A4. Design constraints to extract signal components from the signal to be decomposed in a form that best meets the constraints, thereby achieving signal decomposition; assuming the signal sampling time is: ; The constraints are as follows: ; ; ; ; ; ; in, This represents the total number of samples. For the first The signal components of the first mode of the channel, For demodulated signals, It is a second-order difference operator. It is a norm 2; A5. The instantaneous frequency is updated using arctangent demodulation technology. The frequency update formula is: ; in, For the first The first update Channel 1 The instantaneous frequency of the modal signal components, It is the identity matrix. For the first The first update Channel 1 The instantaneous frequency increment of the modal signal component, , For the first The demodulated signal at the next update Channel 1 Demodulated signal components of the first mode; To simplify the multi-channel input signal model, we assume that all channels have a common frequency component, then: ; in, No. The first update Modal multi-channel common instantaneous frequency components, For the first The first update Channel 1 Instantaneous amplitude of the mode; A6. Update the decomposed signal: ; in, For the first The demodulated signal of the next update. For the first The updated matrix ; A7. Repeat steps A5 to A6 until the stopping condition is met. The stopping condition is reaching the maximum number of updates. In this embodiment, the maximum number of updates is 100.

[0027] S4. To accurately evaluate the effect of signal decomposition, this invention abandons the traditional multi-index combination method and proposes a more robust single objective function based on time-frequency energy separation. The core idea of ​​this method is that a high-quality signal decomposition should result in a highly concentrated energy distribution in the time-frequency domain after the superposition of all decomposed modes, rather than being diffuse like noise. A time-frequency energy distribution matrix is ​​constructed based on the decomposed signal components, and the energy concentration of the energy distribution matrix is ​​evaluated using an objective function. The specific process is as follows: B1. Perform Discrete Short-Time Fourier Transform on the signal components: ; in, For the first The first sampling Channel 1 Modal signal components, The total number of samples, For the first Channel 1 Short-time Fourier transform of modal signal components, For window functions, For time frame indexing, Frequency point index; B2. The amplitude energy spectrum of the signal components is calculated as follows: ; in, For the first Channel 1 Time-frequency energy spectrum of modal signal components, for The complex conjugate; B3. By linearly superimposing the time-frequency energy spectra of all components, the time-frequency energy distribution matrix is ​​obtained as follows: ; in, This is the time-frequency energy distribution matrix; B4. Arrange the elements in the time-frequency energy distribution matrix in ascending order and convert them into a one-dimensional vector: ; in, This represents the total number of elements in the time-frequency energy distribution matrix; B5. Calculate the cumulative energy and total energy of the vector: ; in, For the front The cumulative energy of each element, It is the total energy; At the same time, regulations =0; B6. Calculate the Gini coefficient based on the cumulative energy and the total energy. : ; For the Gini coefficient, the closer the value is to 1, the more concentrated and "unequal" the energy distribution is, that is, the purer the decomposed modes and the higher the separation degree; when the total energy is 0, the Gini coefficient is defined as 1. B7. Express the objective function using the Gini coefficient: ; in, The fitness value is the Gini coefficient; according to the characteristics of the Gini coefficient, the smaller the fitness value, the better the effect.

[0028] S5. Iterate based on the fitness value. In each iteration, update the position of the entire population using an update rule. The iteration process is as follows: C1. Update the positions of the rolling individual, the brooding individual, the offspring individual, and the stolen individual; For a rolling ball individual, the updated position is calculated using the variable spiral greedy selection method and the direct optimization movement method, respectively, and the position with the lower fitness value is taken as the position update value. The core of the variable spiral greedy selection method is a variable number of iterations. Dynamically changing spiral shape parameters : ; in, This represents the maximum number of iterations. The formula for updating an individual's location is: ; ; in, In the variable spiral greedy selection method, the first The individual The individual position in the next iteration It is a random number within the interval [−1, 1]; The direct optimization method updates the individual's position as follows: ; in, Direct Optimization Move Method The individual The individual position in the next iteration and These are the step size control coefficients, which are preset positive constants; For brooding individuals, the location update process is as follows: ; ; in, For the number of iterations, For the first The individual The position of the next iteration. , It is a random number. , These are the lower and upper boundaries of the egg cell's boundary. As the impact factor, , These are the lower and upper bounds of the parameter search space; For impact factors Its expression is: ; in, Maximum number of iterations For child individuals, the position update process is as follows: ; ; in, It is a random number that follows a normal distribution. It is a random number with a value in the range [0,1]. , These are the lower and upper bounds of the sub-individual's position, respectively; For the individual who commits the theft, the location update process is as follows: ; in, As a constant, It is a random value that follows a normal distribution; C2. Correct the updated positions of individuals to ensure that each individual is within the boundary range. The correction process is as follows: when season ; when season ; C3. Using the corrected position as the weight coefficient and smoothness control parameter, execute steps S3~S4 to calculate the fitness value of each individual. C4. Update the historical best position and global best position of the individual based on the fitness value. The update process is as follows: For individuals Compare its first Fitness value of the next iteration The size of its fitness value relative to its historical best position, when season ; Find the individual with the smallest fitness value corresponding to the historical best position. Compare the individual's fitness value with the fitness value of the global optimal position. ,make .

[0029] S6. When the maximum number of iterations is reached, the individual with the lowest fitness value in the population is the globally optimal combination of parameters to be optimized. S7. Substitute the found optimal parameters into the JMCMD method to decompose the original noisy signal. By analyzing the permutation entropy or energy of each mode, identify and remove the noise modes, and superimpose the remaining effective signal modes to obtain a high signal-to-noise ratio signal after noise reduction. Example 2 To fully verify the advancement and effectiveness of the signal decomposition parameter adaptive optimization noise reduction method based on the enhanced dung beetle optimization algorithm proposed in this invention, the following simulated signal experiments and real data experiments are designed.

[0030] 1. Simulated signal experiment (1) Experimental objective This experiment aims to quantitatively verify the performance of the present invention under known signal models and noise levels. Since the clean signal of the simulation signal (i.e., the "standard answer") is known, this experiment can accurately evaluate the invention from the following two aspects: verifying the advantages of the proposed EDBO algorithm in terms of optimization capability and convergence performance compared to the standard DBO algorithm; and demonstrating that automatically optimizing parameters using an optimization algorithm is significantly better than the traditional method of relying on manual experience to set parameters.

[0031] (2) Experimental setup Simulation Signal Construction: A multi-component amplitude-frequency modulation (AM-FM) signal with two channels is constructed to simulate complex vibration signals commonly found in rotating machinery. To simulate a realistic signal, strong Gaussian white noise with SNRs of 3.56 dB and 0 dB is artificially added to the two signal channels. The signal formula is as follows:

[0032] The sampling frequency is 1500Hz, and the signal length is 1500. Each channel contains multiple modal components with similar frequencies that vary over time. Gaussian white noise with a preset signal-to-noise ratio (SNR) is added to the generated clean signal to generate the final noisy simulation signal, which is used for testing all methods.

[0033] The comparison method is as follows.

[0034] The method of this invention (EDBO-JMCMD): Employing the EDBO algorithm described in this invention, which integrates chaotic initialization, nonlinear R-factor, and adaptive foraging strategy, for JMCMD... The parameters were optimized.

[0035] Comparison Method 1 (Standard DBO-JMCMD): The JMCMD parameters are optimized using the standard DBO algorithm without any modifications to highlight the value of the improvements made to the optimization algorithm itself in this invention.

[0036] Comparison method (empirical parameters JMCMD): Without using optimization algorithms, a fixed set of parameters, manually set based on experience, is directly employed. The JMCMD algorithm is run with parameters to demonstrate the necessity of adaptive parameter optimization.

[0037] The evaluation indicators are as follows.

[0038] Since the purity of the signal is known, the following two gold standard quantitative indicators are used for evaluation: Signal-to-noise ratio (SNR): Measures how similar the denoised signal is to the clean signal. A higher SNR value indicates better noise reduction and less signal distortion.

[0039] Root Mean Square Error (RMSE): Measures the absolute error between the denoised signal and the clean signal. The lower the RMSE value, the higher the denoising accuracy.

[0040] (3) Experimental steps Generate a preset noisy multi-channel simulation signal and its corresponding clean signal.

[0041] The noisy signal was processed using the three methods described above, resulting in three sets of denoised signals.

[0042] For methods 1 and 2, record the convergence curves of their optimization process and the optimal parameters found.

[0043] To verify the beneficial effects of the parameter adaptive optimization method based on the improved mantis optimization algorithm proposed in this invention, multiple independent and repeated simulation signal experiments were designed. The experimental results are shown in Table 1 and... Figure 2 The results, as shown, fully demonstrate the advancement and superiority of this invention. The evaluation indicators of all methods are compiled into a table, as shown in Table 1, for visual comparison.

[0044] Table 1 Evaluation metrics for all methods in simulated signals

[0045] First, the parameter adaptive optimization framework proposed in this invention significantly outperforms traditional manual empirical setting methods. As shown in Table 1, after parameter optimization using EDBO or standard DBO, the average signal-to-noise ratio (SNR) (approximately 17.34 dB) is improved by nearly 4 dB compared to the empirical parameter method (13.77 dB), while the average root mean square error is also significantly reduced. This strongly demonstrates the fundamental advantage of the parameter adaptive optimization approach in improving signal decomposition and noise reduction accuracy.

[0046] Secondly, the improvements made to the standard DBO algorithm in this invention result in a significant enhancement of core performance, especially in terms of stability and reliability. Table 1 shows the following key data: ① In terms of stability, the EDBO algorithm of this invention exhibits an overwhelming advantage. Although the average fitness values ​​of the two optimization algorithms are extremely close, the standard deviation of the EDBO algorithm (8.13e-13) is more than an order of magnitude lower than that of the standard DBO algorithm (1.4e-12). This means that in multiple independent experiments, the method of this invention can consistently converge to a high-quality solution range, demonstrating strong robustness and reproducibility, effectively overcoming the large fluctuations in the results of the standard stochastic algorithm. ② In terms of optimization accuracy, the EDBO algorithm of this invention also excels. Although the optimization terrain of the simulation signal is relatively flat, allowing both algorithms to find high-quality solutions in "plain areas," resulting in only slight differences in final performance indicators, EDBO still slightly outperforms in indicators such as average signal-to-noise ratio.

[0047] Finally, statistical tests provide scientific evidence for the superiority of the method of this invention. A Wilcoxon rank-sum test was performed on the final fitness values ​​from 10 independent runs, and the result showed a p-value of 0.0036, far less than the significance level of 0.05. This indicates that the performance difference between the EDBO of this invention and the standard DBO is systematic and statistically significant, rather than due to accidental random fluctuations.

[0048] Appendix Figure 2 This conclusion is presented visually. Figure 2 In the diagram, the average convergence curve (solid red line) of the EDBO algorithm in this invention is enclosed by an extremely narrow shaded area of ​​standard deviation, demonstrating its excellent convergence stability. In contrast, the shaded area of ​​the standard DBO algorithm is relatively wider, exposing the uncertainty of its results. In summary, this invention, through multi-dimensional innovative improvements to the DBO algorithm, greatly enhances the algorithm's stability and reliability while ensuring high optimization accuracy, making it more valuable for solving practical engineering problems.

[0049] 2. Real data experiment (1) Purpose of the experiment This experiment aims to verify the practicality and effectiveness of the invention in real industrial scenarios. Since the intrinsic modes and noise levels of the real signal are unknown, this experiment focuses on evaluating the quality of the decomposition effect.

[0050] (2) Experimental setup First, we introduce the source of the actual data. Vibration signals collected from the TN8000 hydroelectric generator condition monitoring and analysis system are used. This study selects the horizontal vibration signal in the X direction of the stator frame and the horizontal vibration signals of the three stator cores for analysis.

[0051] The evaluation metrics are as follows. Since there is no clean signal as a reference, the Fourier transform spectra of each mode in each channel after decomposition are used to evaluate the advantages of the method. Specifically, the sharper the spectral peaks and the fewer the sidelobes and out-of-band noise, the more thorough the mode separation and the better the effect.

[0052] (3) Experimental steps 1. Load and preprocess the selected 4-channel real hydroelectric generator signal. Noise was added to all four channels. ,in . The values ​​are: 150 for the first channel and 35 for the remaining channels. After adding noise, the signal-to-noise ratios of the four channels are -11.51 dB, -10.27 dB, -11.75 dB and -9.87 dB, respectively.

[0053] 2. For example Figure 3 As shown in the figure, an FFT transformation is performed on the noisy signal.

[0054] 3. The proposed method is used to process the real signal to obtain the decomposed mode set.

[0055] 4. For example Figure 4 As shown, the modes in each channel are subjected to FFT transformation, and their spectra are plotted.

[0056] In terms of quantitative indicators, the method of this invention converges to the lowest final fitness value, proving that it has found a parameter solution with better decomposition quality. In qualitative analysis, the modes decomposed by the method of this invention exhibit the purest single-frequency or narrowband characteristics in their spectra, with the slightest energy leakage and mode aliasing. In summary, this invention, compared with existing technologies, can achieve more thorough signal decomposition and more effective feature extraction when processing real industrial signals with unknown components and low signal-to-noise ratios, and has high engineering application value. Typically, excessive stator core vibration manifests as 1st, 2nd, and 3rd harmonic components, and some components can also be seen in… Figure 3However, in high-noise environments, the 1st, 2nd, and 3rd harmonic frequencies in the spectrum are difficult to distinguish, as noise introduces other high-frequency components that may affect fault determination. The proposed method can accurately extract all spectral components from all channels. Clearly, even operating in high-noise environments, the proposed method can still accurately identify the frequency components in the signal. Therefore, this method can more accurately identify fault characteristics when the stator core vibration of the unit increases. This capability is of great value for diagnosing mechanical faults in noisy environments.

[0057] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A multi-channel signal decomposition parameter self-adaptive optimization denoising method based on an enhanced melolontha optimization algorithm, characterized in that: Includes the following steps: S1, design an enhanced melolontha optimization algorithm to optimize the weight coefficient in the joint extraction-based multi-frequency modulation mode decomposition (JMCMD) method and smoothness control parameters optimization; S2. Initialize the enhanced dung beetle optimization algorithm to generate an initial population. The position of each individual in the population is the initial value of the parameter to be optimized. Initialization includes setting the population size, maximum number of iterations, and search space for the parameters to be optimized in the enhanced dung beetle optimization algorithm; S3. Substitute the initial values ​​of the weighting coefficients and smoothness control parameters into the JMCMD method to decompose the original noisy signal; S4. Construct a time-frequency energy distribution matrix based on the decomposed signal components, and use an objective function to evaluate the energy concentration of the energy distribution matrix; The result of the objective function is expressed in the form of fitness values; S5. Iterate based on fitness values. In each iteration, use update rules to update the position of the entire population. S6. When the maximum number of iterations is reached, the individual with the lowest fitness value in the population is the globally optimal combination of parameters to be optimized. S7. Substitute the found optimal parameters into the JMCMD method to decompose the original noisy signal. By analyzing the permutation entropy or energy of each mode, identify and remove the noise modes, and superimpose the remaining effective signal modes to obtain a high signal-to-noise ratio signal after noise reduction.

2. The noise reduction method of claim 1, wherein: The process of decomposing a signal using the JMCMD method is as follows: S201. Modeling of multivariable linear frequency modulation modes: （1）； wherein is the signal to be decomposed, is the number of signal channels, , , denote the instantaneous amplitude, the instantaneous frequency and the initial phase of the first channel, respectively. S202. Treating the multi-channel signal as a whole to simplify the model, equation (1) can be rewritten as: （2）； （3）； （4）； in, It is white noise and follows a normal distribution. , For the first The instantaneous frequency and demodulation frequency of the first mode. and For the first The demodulated signal components of the first mode, and For the demodulated signal, the first Channel 1 The demodulated signal components of the first mode, The total number of modes, Indicates the first Class modality; S203. Obtain the initial value of the instantaneous frequency of the signal; S204. Design constraints to extract signal components from the signal to be decomposed in the form that best meets the constraints, thereby achieving signal decomposition; assuming the signal sampling time is: （5）； The constraints are as follows: （6）； （7）； （8）； （9）； （10）； （11）； wherein, is the total number of samples, is the number of samples in the first channel, is the signal component of the first modality of the channel, is the demodulated signal, is the second order difference operator, is the two-norm; S205. The instantaneous frequency is updated using arctangent demodulation technology. The frequency update formula is: （12）； wherein is the is the is the is the instantaneous frequency of the modal signal component of the is the identity matrix, is the is the is the instantaneous frequency increment of the modal signal component of the , , is the is the demodulated signal component of the is the order mode of the demodulated signal of the To simplify the multi-channel input signal model, we assume that all channels have a common frequency component, then: （13）； wherein, the first the first modal multi-channel common instantaneous frequency component, the first the first channel the modal instantaneous amplitude; S206. Update the decomposed signal: （14）； wherein is the demodulated signal of the th update, is the matrix of the th update ; S207. Repeat steps S205 to S206 until the stopping condition is met.

3. The noise reduction method of claim 2, wherein: The detailed process of evaluating the objective function is as follows: S301. Perform discrete short-time Fourier transform on the signal components: （15）； wherein is the number of channels, is the number of modalities, is the number of samples, is the number of signal components, is the total number of samples, is the number of channels, is the number of modalities, is the short-time Fourier transform of the signal component of the channel of the modality, is a window function, is a time frame index, is a frequency bin index. S302. Calculate the amplitude energy spectrum of the signal components as follows: （16）； wherein is the first channel the first time-frequency energy spectrum of the modal signal component, is the complex conjugate of ​ S303. By linearly superimposing the time-frequency energy spectra of all components, the time-frequency energy distribution matrix is ​​obtained as follows: （17）； wherein is a time-frequency energy distribution matrix, is the total number of channels, is the total number of modes; S304. Arrange the elements in the time-frequency energy distribution matrix in ascending order and convert them into a one-dimensional vector: （18）； wherein N is the total number of elements in the time-frequency energy distribution matrix; S305. Calculate the cumulative energy and total energy of the vector: （19）； wherein, is the cumulative energy of the preceding elements, is the sum of energies; At the same time, the provisions is 0; S306、According to the accumulated energy and the energy sum, the Gini coefficient is calculated : （20）； When the total energy is 0, the Gini coefficient is defined as 1; S307. Use the Gini coefficient to represent the objective function: （21）； wherein, is a fitness value.

4. The noise reduction method of claim 1, wherein: The enhanced dung beetle optimization algorithm is initialized using an improved Circle chaotic map, as shown in the following formula: （22）； wherein, is the first chaotic value produced by the first iteration, is the current chaotic value, denotes a key control parameter, denotes a control parameter, the function denotes a remainder operation, denotes the control parameter of the improvement part.

5. The noise reduction method of claim 4, wherein: The initial population includes, proportionally, individuals that roll the ball, individuals that breed, offspring, and individuals that steal. For each individual in the initial population set its current position as the individual's historical best position ; for all individuals, find the individual whose historical best position corresponds to the minimum fitness, set the individual's historical best position as the global best position .

6. The noise reduction method of claim 5, wherein: The iterative process is as follows: S601. Update the positions of the rolling individual, the brooding individual, the offspring individual, and the stolen individual; For brooding individuals, the location update process is as follows: （23）； （24）； wherein, is the number of iterations, is the position of the th individual at the th iteration, , is a random number, , is the lower and upper bound of the ovoid boundary, is the impact factor, , is the lower and upper bound of the parameter search space; For child individuals, the position update process is as follows: （25）； （26）； wherein, is a random number subject to a normal distribution, is a random number with a value range of [0, 1], , are lower and upper bounds of the sub-individual position, respectively. For the individual who commits the theft, the location update process is as follows: （27）； wherein, is a constant, is a random value subject to a normal distribution; S602. Correct the updated position of each individual to ensure that each individual is within the boundary range. The correction process is as follows: When time, cause ; When time, cause ; S603. Using the corrected position as the weight coefficient and smoothness control parameter, execute steps S3~S4 to calculate the fitness value of each individual. S604. Update the historical best position and global best position of the individual based on the fitness value. The update process is as follows: For the individual , compare the fitness value of its second iteration to the magnitude of its historical best position fitness value, and when , let ; Find the individual with the smallest fitness value corresponding to the historical best position. Compare the individual's fitness value with the fitness value of the global optimal position. ,make .

7. The noise reduction method according to claim 6, characterized in that: The updated position of the rolling ball is calculated using both the variable spiral greedy selection method and the direct optimization movement method, and the position with the lower fitness value is taken as the position update value.

8. The noise reduction method according to claim 7, characterized in that: The core of the variable spiral greedy selection method is a selection mechanism that varies with the number of iterations. Dynamically changing spiral shape parameters : （28）； in, This represents the maximum number of iterations. The formula for updating an individual's location is: （29）； （30）； in, In the variable spiral greedy selection method, the first The individual The individual position in the next iteration It is a random number within the interval [−1, 1].

9. The noise reduction method according to claim 7, characterized in that: The direct optimization movement method updates the individual position as follows: （31）； in, Direct Optimization Move Method The individual The individual position in the next iteration and These are the step size control coefficients, which are preset normal values.

10. The noise reduction method according to claim 6, characterized in that: The influencing factors The expression is: （32）； in, This represents the maximum number of iterations.

Images