A method and system for online monitoring of gas turbine generator set failure

Abstract

The present application belongs to the technical field of generator set fault monitoring, in order to solve the problem that the traditional technology selects the decomposition parameter of integrated empirical mode decomposition by artificial experience, thus causing mode aliasing, resulting in poor noise reduction effect, and further affecting the accuracy of fault identification, the present application discloses a gas turbine generator set fault online monitoring method and system, and proposes a parameter optimization algorithm with minimum envelope entropy as the optimization target to replace the experience selection, so as to optimize the optimal decomposition parameter; in this way, the problem of mode aliasing existing in the traditional technology can be avoided, thereby improving the noise reduction effect; then, the present application uses the signal distribution trend characteristics to screen out the effective components and reconstruct the signal, so as to obtain the reconstructed vibration signal; finally, the fault feature extraction is carried out on the reconstructed vibration signal, and the reconstructed vibration signal is input into the fault monitoring model, so as to obtain the operation fault monitoring result; thus, the present application ensures the accuracy of the generator set fault identification.

Description

Technical Field

[0001] This invention belongs to the field of generator set fault monitoring technology, specifically relating to an online fault monitoring method and system for gas turbine generator sets. Background Technology

[0002] As the core equipment of a power plant, the operating status of a gas turbine generator set affects the entire power generation process. Due to its complex structure and the close connection between its components, a failure in one part can trigger a chain reaction, causing the entire unit and even the entire power generation process to malfunction. Therefore, in order to ensure the safe operation of the generator set and improve the power supply quality, it is essential to carry out online fault monitoring of the gas turbine generator set.

[0003] Currently, scholars at home and abroad have conducted a lot of research on generator set fault identification. Among them, the most common method is to collect vibration signals during generator set operation, extract fault features, and input the fault features into the fault detection model for fault identification. However, due to the complex operating environment of generator sets, the collected vibration signals often contain a lot of noise. Therefore, before performing fault identification, existing technologies usually use empirical mode decomposition to denoise the signal, thereby reducing the interference of noise on fault identification.

[0004] In practical applications, Empirical Mode Decomposition (EEMD) avoids the difficulty of selecting wavelet bases in wavelet analysis and is more effective than wavelet methods in denoising non-stationary signal decomposition and reconstruction. However, it suffers from mode aliasing, which distorts the decomposed signal components and affects the denoising effect. To address this issue, some researchers have applied the Ensemble Empirical Mode Decomposition (EEMD) algorithm to denoise non-stationary signals by adding white noise. While this method achieves significant denoising, the parameters of the introduced white noise (such as the amplitude coefficient) and the number of average processing iterations significantly influence the results. Traditional techniques typically rely on empirical selection, and improper parameter selection can still lead to mode aliasing, resulting in poor denoising and affecting the accuracy of subsequent generator fault identification. Therefore, given these shortcomings, providing a gas turbine generator fault online monitoring method with good denoising performance that improves the accuracy of generator fault identification has become an urgent problem to solve. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for online fault monitoring of gas turbine generator sets, which solves the problem that the technology of using manual experience to select decomposition parameters for integrated empirical mode decomposition to decompose vibration signals leads to mode aliasing, resulting in poor noise reduction effect and thus affecting the accuracy of fault identification.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] Firstly, a method for online fault monitoring of gas turbine generator sets is provided, including:

[0008] Acquire operating vibration signals of the gas turbine generator set;

[0009] With the goal of minimizing the envelope entropy of signal decomposition, the operational vibration signal is subjected to ensemble empirical mode decomposition parameter optimization processing to obtain the optimal decomposition parameters of the operational vibration signal.

[0010] Based on the optimal decomposition parameters, the running vibration signal is subjected to ensemble empirical mode decomposition to obtain several component signals.

[0011] The signal distribution trend characteristics between each component signal and the running vibration signal are calculated, and the effective component is determined from several component signals based on the signal distribution trend characteristics between each component signal and the running vibration signal.

[0012] The effective components are reconstructed to obtain a reconstructed vibration signal, and the reconstructed vibration signal is processed by feature extraction to obtain the fault characteristics of the gas turbine generator set.

[0013] The fault characteristics are input into the fault monitoring model to obtain the operational fault monitoring results of the gas turbine generator set.

[0014] Based on the above disclosure, after acquiring the operating vibration signal of the gas turbine generator set, this invention optimizes the ensemble empirical mode decomposition parameters by using the minimum envelope entropy of signal decomposition as the optimization objective, thereby obtaining the optimal decomposition parameters of the operating vibration signal. Then, based on the aforementioned optimal decomposition parameters, the operating vibration signal is processed by ensemble empirical mode decomposition to obtain several component signals. Next, the signal distribution trend characteristics between each component signal and the operating vibration signal are calculated, and based on this, signal filtering is performed, that is, the effective components are determined from the several component signals. Then, the effective components are reconstructed to obtain the reconstructed vibration signal, thereby completing signal denoising. Finally, the fault features of the reconstructed vibration signal are extracted and input into the fault monitoring model to obtain the operating fault monitoring results of the generator set.

[0015] Through the above design, this invention proposes a parameter optimization algorithm with minimum envelope entropy as the optimization objective, replacing manual experience in parameter selection to find the optimal decomposition parameters for ensemble empirical mode decomposition. This avoids the mode aliasing problem inherent in traditional techniques that rely on manual experience in parameter selection, thus improving denoising performance. Furthermore, this invention utilizes the signal distribution trend characteristics between the components and the original signal to filter out effective components and reconstruct the signal, obtaining a reconstructed vibration signal. Finally, fault features are extracted from the reconstructed vibration signal and input into a fault monitoring model to obtain the generator set's operational fault monitoring results. Therefore, this invention improves the noise reduction effect of generator set vibration signals, ensuring the accuracy of generator set fault identification, and is highly suitable for large-scale application and promotion.

[0016] In one possible design, with the goal of minimizing the envelope entropy of the signal decomposition, the ensemble empirical mode decomposition parameters of the running vibration signal are optimized to obtain the optimal decomposition parameters of the running vibration signal, including:

[0017] The decomposition parameters of the set empirical mode decomposition are initialized to obtain multiple sets of decomposition parameters, and an optimal initial population is constructed based on the multiple sets of decomposition parameters, wherein each individual in the optimal initial population corresponds to a set of decomposition parameters.

[0018] Obtain the individual population at the nth iteration and calculate the fitness of each individual in the individual population at the nth iteration. The fitness of any individual is the minimum envelope entropy of several component signals obtained by performing ensemble empirical mode decomposition on the running vibration signal based on the decomposition parameters corresponding to that individual. The initial value of n is 1, and when n is 1, the individual population at the nth iteration is the optimal initial population.

[0019] Based on the fitness of each individual, determine whether the population of individuals is in a globally convergent state at the nth iteration;

[0020] If not, then perform a global search on the individual population at the nth iteration to obtain the individual population at the (n+1)th iteration;

[0021] Increment n by 1 and reacquire the individual population at the nth iteration until the individual population at the nth iteration is in a globally convergent state, thus obtaining the globally optimal population.

[0022] Using the minimum envelope entropy of signal decomposition as the search objective, the global optimal population is subjected to iterative local search processing until the iteration meets the stopping condition, and the global optimal individual is obtained.

[0023] Based on the globally optimal individual, the optimal decomposition parameters are determined.

[0024] In one possible design, a global search is performed on the population of individuals at the nth iteration to obtain the population of individuals at the (n+1)th iteration, including:

[0025] The position of each individual in the population at the nth iteration is perturbed once to obtain the perturbed population;

[0026] The global search evolution value of the perturbed population is calculated based on the current iteration number and the maximum iteration number of the global search.

[0027] Based on the global search evolution value, a global search process is performed on the disturbed population to obtain a new individual population;

[0028] Based on the fitness of each updated individual in the new individual population, a second position perturbation process is performed on each updated individual to obtain the individual population at the (n+1)th iteration after the second position perturbation process.

[0029] In one possible design, the position of each individual in the population at the nth iteration is perturbed once to obtain a perturbed population, including:

[0030] For any individual in the population at the nth iteration, a target individual is selected from the population at the nth iteration, wherein the target individual is different from any individual.

[0031] Calculate the first distance between any individual and the best individual, and the second distance between the target individual and the best individual, wherein the best individual is the individual with the smallest fitness in the individual population at the nth iteration;

[0032] The first disturbance control factor is calculated based on the first distance and the second distance;

[0033] Obtain the global search range of any individual, and calculate the first perturbation intermediate parameter based on the global search range, the first distance, and the individual; and calculate the second perturbation intermediate parameter based on the global search range, the second distance, and the target individual.

[0034] The second disturbance control factor is calculated based on the first and second disturbance intermediate parameters.

[0035] Based on the first perturbation control factor, the second perturbation control factor, and the global search range, a position perturbation process is performed on any individual to obtain the perturbed individual corresponding to any individual. After all individuals in the individual population at the nth iteration have been polled, the perturbed individuals corresponding to each individual are used to form the perturbed population.

[0036] In one possible design, the global search evolution value of the perturbed population is calculated based on the current iteration number and the maximum iteration number of the global search, including:

[0037] Calculate the ratio between the current iteration number and the maximum iteration number, and calculate the difference between the ratio threshold and the ratio to obtain the evolution coefficient;

[0038] Obtain the first evolutionary random number, and determine the global search evolutionary value based on the evolutionary coefficient and the first evolutionary random number;

[0039] Accordingly, based on the global search evolution value, a global search is performed on the perturbed population to obtain a new individual population, which includes:

[0040] Determine whether the global search evolution value is greater than the first evolution threshold;

[0041] If not, then determine whether the global search evolution value is greater than or equal to the second evolution threshold, wherein the second evolution threshold is less than the first evolution threshold;

[0042] If so, then generate a second evolutionary random number;

[0043] Determine whether the second evolutionary random number is less than the second evolutionary threshold;

[0044] If not, then a global search coefficient is generated, and the best perturbation individual is selected from the perturbation population. Based on the best perturbation individual, the global search coefficient, and the global search evolution value, a global search process is performed on each perturbation individual in the perturbation population to obtain the updated individual corresponding to each perturbation individual, so as to form a new individual population using each updated individual.

[0045] In one possible design, if the global search evolution value is greater than a first evolution threshold, the method further includes:

[0046] Generate several first position random numbers for each disturbed individual in the disturbed population;

[0047] Calculate the average position of individuals in the disturbed population;

[0048] Based on several first position random numbers, the average position of the individuals, and the best perturbation individual, a global search process is performed on each perturbation individual to obtain the updated individual corresponding to each perturbation individual;

[0049] If the global search evolution value is less than the second evolution threshold, the method further includes:

[0050] Generate a third evolutionary random number and determine whether the third evolutionary random number is greater than or equal to the second evolutionary threshold;

[0051] If so, then based on the optimal perturbation individual and the global search evolution value, perform a global search process on each perturbation individual to obtain the updated individual corresponding to each perturbation individual; otherwise, based on the average position of the individual, the optimal perturbation individual, the global search coefficient, and the global search evolution value, perform a global search process on each perturbation individual to obtain the updated individual corresponding to each perturbation individual.

[0052] In one possible design, based on the fitness of each updated individual in the new individual population, a secondary position perturbation process is applied to each updated individual, including:

[0053] For any updated individual in the new individual population, the perturbation intensity is calculated based on the fitness of that updated individual;

[0054] Generate a random number for the second position;

[0055] Determine whether the random number at the second position is less than the perturbation probability;

[0056] If so, then from the new individual population, determine the updated individual corresponding to the smallest fitness as the designated individual, and perform a secondary position perturbation process on any updated individual according to the second position random number, the perturbation intensity and the designated individual to obtain the secondary perturbation individual corresponding to any updated individual;

[0057] Otherwise, two updated individuals are randomly selected from the new individual population as the first random individual and the second random individual. Based on the first random individual, the second random individual, the second position random number, and the perturbation intensity, a secondary position perturbation process is performed on any of the updated individuals to obtain the secondary perturbation individual corresponding to any of the updated individuals.

[0058] In one possible design, the signal distribution trend characteristics between each component signal and the operating vibration signal are calculated, including:

[0059] For any component signal, calculate the first mean of the sorting indices of all signal values ​​in the component signal and the second mean of the sorting indices of all signal values ​​in the running vibration signal, wherein the sorting index of any signal value in the component signal is the sorting sequence number of the signal value in the component signal.

[0060] The difference between the sorting index of each signal value in any component signal and the first mean is calculated to obtain several first sorting index differences; and the difference between the sorting index of each signal value in the running vibration signal and the second mean is calculated to obtain several second sorting index differences.

[0061] Based on several first sorting index differences and several second sorting index differences, the distribution difference degree and distribution dispersion degree between any component signal and the running vibration signal are calculated respectively.

[0062] The signal distribution trend characteristics are calculated based on the distribution difference and distribution dispersion.

[0063] In one possible design, based on the signal distribution trend characteristics between each component signal and the operating vibration signal, the effective components are determined from several component signals, including:

[0064] For any component signal, determine the fluctuation trend characteristics between the any component signal and the operating vibration signal;

[0065] Calculate the amplitude difference between any component signal and the running vibration signal;

[0066] Based on the fluctuation trend characteristics, the amplitude difference degree, and the signal distribution trend characteristics, the correlation between any component signal and the operating vibration signal is calculated, and after polling all component signals, the correlation between each component signal and the operating vibration signal is obtained.

[0067] Based on the correlation between each component signal and the operating vibration signal, the effective component is determined from several component signals.

[0068] Secondly, an online fault monitoring system for gas turbine generator sets is provided, including:

[0069] The acquisition unit is used to acquire the operating vibration signals of the gas turbine generator set;

[0070] The parameter optimization unit is used to perform ensemble empirical mode decomposition parameter optimization processing on the running vibration signal with the goal of minimizing the envelope entropy of signal decomposition, so as to obtain the optimal decomposition parameters of the running vibration signal.

[0071] The signal decomposition unit is used to perform ensemble empirical mode decomposition on the running vibration signal based on the optimal decomposition parameters to obtain several component signals.

[0072] The signal reconstruction unit is used to calculate the signal distribution trend characteristics between each component signal and the running vibration signal, and to determine the effective component from several component signals based on the signal distribution trend characteristics between each component signal and the running vibration signal.

[0073] The signal reconstruction unit is also used to reconstruct the effective components to obtain a reconstructed vibration signal, and to perform feature extraction processing on the reconstructed vibration signal to obtain the fault characteristics of the gas turbine generator set.

[0074] The fault monitoring unit is used to input the fault characteristics into the fault monitoring model to obtain the operational fault monitoring results of the gas turbine generator set.

[0075] Thirdly, an online fault monitoring device for a gas turbine generator set is provided. Taking the device as an electronic device as an example, it includes a memory, a processor, and a transceiver that are connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the online fault monitoring method for the gas turbine generator set as described in the first aspect or any possible design of the first aspect.

[0076] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the online fault monitoring method for gas turbine generator sets as described in the first aspect or any possible design of the first aspect.

[0077] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, causes the computer to perform the online fault monitoring method for gas turbine generator sets as described in the first aspect or any possible design of the first aspect.

[0078] Beneficial effects:

[0079] (1) This invention proposes a parameter optimization algorithm with minimum envelope entropy as the optimization objective, which replaces manual experience selection to find the optimal decomposition parameters of the set experience mode decomposition. In this way, the mode aliasing problem existing in the traditional technology that relies on manual experience to select decomposition parameters can be avoided, thereby improving the denoising effect. Then, this invention uses the signal distribution trend characteristics between the components and the original signal to screen out the effective components and reconstruct the signal to obtain the reconstructed vibration signal. Finally, the fault features of the reconstructed vibration signal are extracted and the extracted fault features are input into the fault monitoring model to obtain the generator set operation fault monitoring results. Thus, this invention can improve the denoising effect of the generator set vibration signal, thereby ensuring the accuracy of generator set fault identification. Therefore, it is very suitable for large-scale application and promotion. Attached Figure Description

[0080] Figure 1 This is a flowchart illustrating the steps of the online fault monitoring method for gas turbine generator sets provided in an embodiment of the present invention.

[0081] Figure 2 This is a schematic diagram of the structure of the online fault monitoring system for gas turbine generator sets provided in an embodiment of the present invention;

[0082] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0083] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0084] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0085] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0086] Example:

[0087] See Figure 1As shown, the online fault monitoring method for gas turbine generator sets provided in this embodiment optimizes the parameters of ensemble empirical mode decomposition by using the minimum envelope entropy as the optimization objective, thereby obtaining the optimal decomposition parameters for the generator set's operating vibration signal. By replacing manual parameter selection with the aforementioned optimization algorithm, the mode aliasing problem inherent in traditional techniques relying on manual parameter selection is avoided, thus improving the denoising effect. After decomposing the vibration signal using the aforementioned optimal decomposition parameters, this method utilizes the signal distribution trend characteristics between the components and the original signal to filter out effective components and reconstruct the signal, thereby obtaining the reconstructed vibration signal. Finally, the reconstructed vibration signal is processed... Fault features are extracted and input into the fault monitoring model to obtain the generator set's operational fault monitoring results. Therefore, this method can improve the noise reduction effect of generator set vibration signals, reduce the interference of noise signals on fault identification, and thus ensure the accuracy of generator set fault identification. For example, this method can be run on the generator set's online monitoring terminal. Optionally, the online monitoring terminal can be, but is not limited to, a personal computer (PC) or a server. It is understood that the aforementioned execution entity does not constitute a limitation on the embodiments of this application. Accordingly, the operation steps of this method can be, but are not limited to, the steps S1 to S6 below.

[0088] S1. Acquire the operating vibration signal of the gas turbine generator set; In this embodiment, for example, but not limited to, vibration acceleration sensors can be installed on different parts of the gas turbine generator set (such as the housing of bearings, rotors, etc.) to collect operating vibration signals from different parts and upload them to the online monitoring terminal. The online monitoring terminal can then perform noise reduction and fault feature extraction on the received operating vibration signals and identify faults in different parts of the gas turbine generator set based on the fault features. Since the generator set usually operates in a complex environment, the collected operating vibration signals will inevitably contain a lot of noise. Therefore, in order to reduce the interference of noise on fault identification, the operating vibration signals need to be noise-reduced before fault feature extraction.

[0089] Furthermore, traditional techniques typically employ ensemble empirical mode decomposition (EMD) algorithms to decompose running vibration signals. Then, effective signals are selected from the decomposed component signals for reconstruction, thereby removing noise from the original signal. However, the amplitude coefficients of the Gaussian white noise added in traditional EMD, as well as the average number of processing iterations in the algorithm, are selected empirically. Therefore, improper parameter selection can also lead to aliasing of component modes. To address these shortcomings, this embodiment proposes a parameter optimization algorithm with minimum envelope entropy as the optimization objective. This algorithm optimizes the decomposition parameters, thereby avoiding the aliasing problem caused by empirically selected decomposition parameters.

[0090] Optionally, the parameter optimization process may be, but is not limited to, the steps shown in S2 below.

[0091] S2. Using the minimum envelope entropy of signal decomposition as the optimization objective, the running vibration signal is subjected to ensemble empirical mode decomposition parameter optimization processing to obtain the optimal decomposition parameters of the running vibration signal. In specific implementation, envelope entropy represents the sparsity characteristics of the original signal. When the component signals corresponding to the original signal have more noise and less feature information, the envelope entropy is larger, and vice versa. Therefore, this embodiment uses the minimum envelope entropy of signal decomposition as the optimization objective to optimize the decomposition parameters, thereby finding a set of parameters that minimizes the envelope entropy of the decomposed signal, thus ensuring that the decomposed signal contains the least amount of noise.

[0092] Optionally, this embodiment provides a parameter optimization algorithm that first performs a global search and then a fast and precise local search. That is, a global search is performed first, and then after the global search converges (i.e., converges to the interval where the global optimum is located), a fast and accurate local search method is used to perform a fast local search in the interval where the global optimum is located, so as to find the global optimum in the interval where the global optimum is located, and then obtain the optimal decomposition parameters. The aforementioned search process is shown in steps S21 to S27 below.

[0093] S21. Initialize the decomposition parameters of the ensemble empirical mode decomposition to obtain multiple sets of decomposition parameters, and construct an optimal initial population based on the multiple sets of decomposition parameters. Each individual in the optimal initial population corresponds to a set of decomposition parameters. In this embodiment, the decomposition parameters may include, but are not limited to, white noise amplitude coefficients and average number of processing times. Thus, the position of each individual corresponds to a set of decomposition parameters. Based on this, by updating the individual positions and using the minimum envelope entropy as the fitness function, the decomposition parameters can be optimized.

[0094] Furthermore, in this embodiment, an initial population is first constructed using multiple sets of decomposition parameters. Then, to ensure the diversity of the initial population, diversification processing is performed on it. The specific process is as follows:

[0095] S21a. Calculate the diversity value of the initial population; in this embodiment, for example, but not limited to, the following formula (1) can be used to calculate the diversity value of the initial population.

[0096] (1)

[0097] In the above formula (1), This represents the diversity value. This represents the value of the d-th dimension in the position vector of the m-th initial individual in the initial population. Let represent the d-th dimension mean of the position vectors of all initial individuals in the initial population. Let represent the maximum value of the d-th dimension in the position vector of all initial individuals in the initial population. Let represent the minimum value of the d-th dimension in the position vector of all initial individuals in the initial population. This represents the total dimension of the position vector. Indicates the size of the initial population, where, .

[0098] S21b. Determine whether the diversity value is less than the diversity threshold; in this embodiment, if the diversity value is less than the diversity threshold, then diversification processing is required, and the process is as shown in step S21c below.

[0099] S21c. Using a population mapping function, generate a mapped population, wherein the size of the mapped population is the same as the size of the initial population.

[0100] In practical implementation, for example, the population mapping function can be set as, but is not limited to:

[0101] (2)

[0102] In equation (2), Let represent the population mapping function, and also the mapping value at the (c+1)th iteration. This represents the mapping value at the c-th iteration.

[0103] Thus, for any initial individual in the initial population, a given initial individual is assigned... The initial value is then used to iterate the aforementioned formula (2). This process is repeated several times to obtain the mapping vector corresponding to any initial individual. Finally, the mapping vector can be used as the mapping individual corresponding to any initial individual.

[0104] After generating the mapping population based on the aforementioned formula (2), the fitness can be calculated, as shown in step S21d below.

[0105] S21d. The initial population and the mapped population are merged to obtain a merged population. The fitness of each merged individual in the merged population is calculated, and the merged individuals are sorted in ascending order of fitness to obtain a sorted sequence. In this embodiment, as previously explained, the position vector of each initial individual is a set of decomposition parameters. Therefore, based on the position vector of the merged individuals, the running vibration signal can be subjected to ensemble empirical mode decomposition to obtain several component signals corresponding to each merged individual. Then, for any merged individual, the envelope entropy of each component signal corresponding to that merged individual is calculated. Next, the sum of the envelope entropies is calculated to obtain the fitness of any merged individual. Thus, the smaller the fitness of a merged individual, the better its position and the more reasonable the corresponding decomposition parameters. Of course, envelope entropy is a commonly used calculation method to measure the sparsity of a signal, and its calculation process will not be elaborated here.

[0106] S21e. Take the first part of the sorted sequence... The merged individuals form the optimal initial population.

[0107] Therefore, through the aforementioned steps S21a to S21e, population diversity is calculated, and when the diversity is below a threshold, individuals are mapped and merged with the original population. Finally, the population with the highest fitness is selected. By using a number of individuals to form the optimal initial population, the diversity of the initial population can be increased, thereby ensuring the accuracy of subsequent searches.

[0108] After obtaining the optimal initial population, a global search can be performed on the optimal initial population, as shown in step S22 below.

[0109] S22. Obtain the individual population at the nth iteration and calculate the fitness of each individual in the individual population at the nth iteration. The fitness of any individual is the minimum envelope entropy of several component signals obtained by performing ensemble empirical mode decomposition on the running vibration signal based on the decomposition parameters corresponding to that individual. The initial value of n is 1, and when n is 1, the individual population at the nth iteration is the optimal initial population. In this embodiment, the fitness calculation process has been described above, and the process will not be repeated here.

[0110] After obtaining the fitness of each individual at the nth iteration, global convergence can be determined, as shown in step S23 below.

[0111] S23. Based on the fitness of each individual, determine whether the population of individuals at the nth iteration is in a globally convergent state. In this embodiment, the standard deviation at the nth iteration is obtained based on the standard deviation of the fitness of the population of individuals at the nth iteration. Then, the standard deviations from the (n-5)th to the (n-1)th iterations are obtained. Next, the change values ​​between the standard deviations from the (n-5)th to the nth iterations are calculated (i.e., the change in the standard deviation between the (n-5)th and (n-4)th iterations, the change in the standard deviation between the (n-4)th and (n-3)th iterations, and so on, up to the change in the standard deviation between the (n-1)th and the nth iterations). Finally, determine whether the change values ​​are all less than the change threshold. If so, it indicates that the population of individuals at the nth iteration is in a globally convergent state; otherwise, it is not in a globally convergent state.

[0112] In this embodiment, if the individual population is not in a globally converged state at the nth iteration, a global search is required, and the process is shown in step S24 below.

[0113] S24. If not, then perform a global search on the individual population at the nth iteration to obtain the individual population at the (n+1)th iteration; in specific implementation, a global search process can be, but is not limited to, the steps S24a to S24d below.

[0114] S24a. Perform a position perturbation process on each individual in the individual population at the nth iteration to obtain a perturbed population. In this embodiment, the position perturbation is performed to further improve the population diversity, thereby preventing the population from getting trapped in local optima during global search. The position perturbation process can be, but is not limited to, as shown in steps S24a1 to S24a6 below.

[0115] S24a1. For any individual in the individual population at the nth iteration, a target individual is selected from the individual population at the nth iteration, wherein the target individual is different from any individual. In this embodiment, after randomly selecting an individual different from any individual from the individual population at the nth iteration as the target individual, the position of any individual can be perturbed based on this, and the process is as shown in the following steps S24a2 to S24a6.

[0116] S24a2. Calculate the first distance between any individual and the best individual, and the second distance between the target individual and the best individual, wherein the best individual is the individual with the smallest fitness in the population of individuals at the nth iteration; in specific implementation, the Euclidean distance between the position vectors of any individual and the best individual can be used as the first distance between them; of course, the calculation method of the second distance is the same, and will not be repeated here.

[0117] After obtaining the distance between any individual, the target individual, and the optimal individual, the first disturbance control factor can be calculated, as shown in step S24a3 below.

[0118] S24a3. Calculate the first disturbance control factor based on the first distance and the second distance; in specific applications, for example, but not limited to, the ratio between the first distance and the second distance can be used as the first disturbance control factor; then, the second disturbance control factor can be calculated, as shown in steps S24a4 and S24a5 below.

[0119] S24a4. Obtain the global search range of any individual, and calculate the first perturbation intermediate parameter based on the global search range, the first distance, and the any individual, and calculate the second perturbation intermediate parameter based on the global search range, the second distance, and the target individual.

[0120] In this embodiment, the global search range includes the search range of each dimension of the position vector of any individual. Therefore, when calculating the first perturbation intermediate parameter, it is necessary to calculate it according to the dimension, and the calculation formula is as follows:

[0121] (3)

[0122] In equation (3), Let represent the first perturbation intermediate parameter corresponding to the d-th dimension in the position vector of any individual. Let represent the maximum and minimum values ​​of the search range in the d-th dimension, respectively. This represents the value of the d-th dimension in the position vector of any individual. This represents the first distance.

[0123] Similarly, for example, but not limited to, the following formula (4) can be used to calculate the intermediate parameters of the second disturbance.

[0124] (4)

[0125] In equation (4), This represents the second perturbation intermediate parameter corresponding to the d-th dimension of the position vector of the target individual corresponding to any given individual. This represents the value of the d-th dimension in the position vector of the target individual. This represents the second distance.

[0126] Thus, by using the aforementioned formulas (3) and (4), the first and second disturbance intermediate parameters of each dimension in the position vector can be calculated; and then, based on this, the second disturbance control factor can be calculated, as shown in step S24a5 below.

[0127] S24a5. Calculate the second disturbance control factor based on the first and second disturbance intermediate parameters. In specific implementation, each dimension of the position vector of any individual corresponds to a second disturbance control factor. The second disturbance control factor corresponding to the d-th dimension is the absolute value of the ratio between the first disturbance intermediate parameter corresponding to the d-th dimension of the position vector of any individual and the second disturbance intermediate parameter corresponding to the d-th dimension of the position vector of the target individual. Thus, after calculating the second disturbance control factor for each dimension, the position disturbance of any individual can be performed, as shown in step S24a6 below.

[0128] S24a6. Based on the first perturbation control factor, the second perturbation control factor, and the global search range, perform a position perturbation process on any individual to obtain the perturbed individual corresponding to any individual. After all individuals in the individual population at the nth iteration have been polled, the perturbed individuals corresponding to each individual are used to form the perturbed population.

[0129] In this embodiment, for the d-th dimension of the position vector of any individual, a position perturbation can be performed using, but is not limited to, the following formula (5).

[0130] (5)

[0131] In the above formula (5), This represents the perturbation value corresponding to the d-th dimension in the position vector of any individual. This represents the value of the d-th dimension in the position vector of any individual. Let represent the first disturbance control factor and the second disturbance control factor in the d-th dimension, respectively.

[0132] Thus, by using the aforementioned formula (5), the perturbation value corresponding to each dimension in the position vector of any individual can be calculated. Then, by using the perturbation value of each dimension, the perturbation individual corresponding to any individual can be formed (of course, it refers to the position vector corresponding to the perturbation individual).

[0133] Therefore, after completing one position perturbation of the individual population in the nth iteration through the aforementioned steps S24a1 to S24a6, position updates can be performed, as shown in steps S24b and S24c below.

[0134] S24b. Calculate the global search evolution value of the perturbed population based on the current iteration number and the maximum iteration number of the global search. In specific implementation, for example, but not limited to, the ratio between the current iteration number and the maximum iteration number can be calculated, and the difference between the ratio threshold and the ratio can be calculated to obtain the evolution coefficient. Then, obtain the first evolution random number. Finally, determine the global search evolution value based on the evolution coefficient and the first evolution random number. In this embodiment, for example, the value range of the first evolution random number is (-1, 1), and the ratio threshold is set to 1. Thus, after calculating the global search evolution value of the aforementioned perturbed population, a global search can be performed on the perturbed population based on this, as shown in step S24c below.

[0135] S24c. Based on the global search evolution value, perform a global search on the perturbed population to obtain a new individual population; in this embodiment, for example, but not limited to, the following steps S24c1 to S24c5 can be used to complete a global search of the perturbed population.

[0136] S24c1. Determine whether the global search evolution value is greater than the first evolution threshold; in specific implementation, for example, the first evolution threshold may be set to 1, but is not limited to; wherein, if the global search evolution value is greater than the first evolution threshold, the following operation is performed:

[0137] Step 1: Generate several first position random numbers for each disturbed individual in the disturbed population; in this embodiment, each disturbed individual corresponds to 4 first position random numbers. After generating the first position random numbers for each disturbed individual, the average position of the individual is calculated, as shown in Step 2 below.

[0138] Step 2: Calculate the average position of individuals in the perturbed population; in this embodiment, the average position of individuals is the mean vector of the position vectors of all individuals in the perturbed population; after obtaining the average position of individuals, the position of each perturbed individual can be updated, and the process is shown in Step 3 below.

[0139] Step 3: Based on several first position random numbers, the average position of the individuals, and the best perturbation individual, perform a global search on each perturbation individual to obtain the updated individuals corresponding to each perturbation individual, so as to form a new individual population using the updated individuals. The best perturbation individual is the perturbation individual with the lowest fitness in the perturbation population.

[0140] In this embodiment, taking any disturbed individual as an example, a third position random number is first generated. Then, the position is updated based on the third position random number, several first position random numbers, the average position of the individual, and the optimal disturbed individual; wherein, the position update formula is:

[0141] (6)

[0142] In equation (6), This represents the updated individual corresponding to any perturbed individual. This represents the optimal perturbation individual. This refers to any of the disturbed individuals. This represents a random number in the first position (with values ​​ranging from 0 to 1). As a third random individual, Indicates the average position of individuals. This represents a random number in the third position (the value range is also between 0 and 1). The lower and upper limits of the search space (the minimum and maximum values ​​of the global search range) are represented, where the third random individual is any perturbation individual in the perturbation population that is different from any of the perturbation individuals.

[0143] Thus, when the global search evolution value is greater than the first evolution threshold, the following formula (6) is used to update the position of each perturbation individual; and when the global search evolution value is less than or equal to the first evolution threshold, the following step S24c2 needs to be executed.

[0144] S24c2. If not, determine whether the global search evolution value is greater than or equal to the second evolution threshold, wherein the second evolution threshold is less than the first evolution threshold; in specific implementation, the second evolution threshold is set to 0.5, wherein when the global search evolution value is greater than 0.5, the following step S24c3 needs to be executed.

[0145] S24c3. If so, generate a second evolutionary random number; in this embodiment, the value range of the second evolutionary random number is (0,1). After obtaining the second evolutionary random number, the second random number can be judged, and the process is as shown in step S24c4 below.

[0146] S24c4. Determine whether the second evolutionary random number is less than the second evolutionary threshold; In this embodiment, when the second evolutionary random number is greater than or equal to the second evolutionary threshold, it is necessary to select the best perturbation individual from the perturbation population so as to combine the best perturbation individual to perform position update. The update process is shown in step S24c5 below.

[0147] S24c5. If not, generate global search coefficients and select the best perturbation individual from the perturbation population. Based on the best perturbation individual, the global search coefficients, and the global search evolution value, perform a global search process on each perturbation individual in the perturbation population to obtain the updated individual corresponding to each perturbation individual, so as to form a new individual population using each updated individual.

[0148] In this embodiment, the position update process of step S24c5 is illustrated by taking any disturbed individual as an example. For example, the updated individual corresponding to any disturbed individual can be obtained by using the following formula (7).

[0149] (7)

[0150] In equation (7), This represents the updated individual corresponding to any perturbed individual. This represents the optimal perturbation individual. This refers to any of the disturbed individuals. Indicates the global search evolution value. This represents the global search coefficient.

[0151] Furthermore, in the aforementioned step S24c4, if the second evolutionary random number is less than the second evolutionary threshold, the following steps are required to complete the position update of each perturbed individual (i.e., a global search).

[0152] Step (1): Based on the best perturbation individual, the global search coefficient and the global search evolution value, perform a pre-global search process on each perturbation individual in the perturbation population to obtain the first pre-updated individual corresponding to each perturbation individual.

[0153] In practical applications, for example, but not limited to, the following formula (8) can be used to perform a pre-global search process on any disturbed individual.

[0154] (8)

[0155] In equation (8), This represents the first pre-updated individual corresponding to any perturbation individual.

[0156] After completing a pre-global search for each disturbed individual, fitness can be determined, as shown in step (2) below.

[0157] Step (2): For any perturbed individual, determine whether the fitness of the perturbed individual is less than the fitness of the first pre-updated individual corresponding to the perturbed individual.

[0158] Step (3): If not, perform a second pre-global search on the first pre-update individual of any disturbed individual to obtain the second pre-update individual corresponding to any disturbed individual; in this embodiment, for example, but not limited to, perform a second pre-global search on the first pre-update individual of any disturbed individual according to the following formula (9).

[0159] (9)

[0160] In equation (9), This represents the second pre-updated individual corresponding to any perturbed individual. for A random vector of dimension 1. This is the Levy flight coefficient.

[0161] Thus, after completing the second pre-global search process of the first pre-update individual for any disturbed individual according to the aforementioned formula (9), fitness can be judged again, as shown in the following step (4).

[0162] Step (4): Determine whether the fitness of the second pre-updated individual corresponding to any of the perturbed individuals is less than the fitness of any of the perturbed individuals.

[0163] Step (5): If yes, then the second pre-updated individual corresponding to any perturbation individual is taken as the updated individual corresponding to any perturbation individual, and after all perturbation individuals have been polled, the updated individuals corresponding to each perturbation individual are obtained; In this embodiment, if the fitness of the second pre-updated individual corresponding to any perturbation individual is less than the fitness of any perturbation individual, it means that the position of the second pre-updated individual is better and needs to accept a new solution; otherwise, if the aforementioned conditions are not met, the original position is maintained, that is, any perturbation individual is directly taken as the updated individual; Similarly, the same applies in step (2), which will not be repeated here.

[0164] Thus, through the aforementioned steps, the position update of each perturbed individual can be completed when the global search evolution value is greater than or equal to the second evolution threshold; simultaneously, when the global search evolution value is less than the second evolution threshold, the following operation is performed:

[0165] Step (6): Generate a third evolutionary random number and determine whether the third evolutionary random number is greater than or equal to the second evolutionary threshold. In this embodiment, the third evolutionary random number is also a random number between (0,1). After obtaining the third evolutionary random number, it can be compared with the second evolutionary threshold so that different position update operations can be performed according to the comparison result. The process is shown in step (7) below.

[0166] Step (7): If yes, then perform a global search process on each perturbation individual based on the best perturbation individual and the global search evolution value to obtain the updated individual corresponding to each perturbation individual; otherwise, perform a global search process on each perturbation individual based on the average position of the individual, the best perturbation individual, the global search coefficient and the global search evolution value to obtain the updated individual corresponding to each perturbation individual.

[0167] In this embodiment, when the third evolutionary random number is greater than or equal to the second evolutionary threshold, a global search process is performed on any disturbed individual according to the following formula (10).

[0168] (10)

[0169] Similarly, when the third evolutionary random number is less than the second evolutionary threshold, a pre-global search is performed on each of the perturbed individuals in the perturbed population based on the average position of the individual, the best perturbed individual, the global search coefficient, and the global search evolutionary value to obtain the first pre-updated individual corresponding to each perturbed individual; then, the aforementioned steps (2) to (5) are used to complete the position update of each perturbed individual, that is, to obtain the updated individual corresponding to each perturbed individual.

[0170] When the third evolutionary random number is greater than the second evolutionary threshold, for example, but not limited to, the following formula (11) can be used to perform a pre-global search process on any of the perturbed individuals.

[0171] (11)

[0172] Thus, through the aforementioned steps S24c1 to S24c5 and their sub-steps, a global search process can be completed for each perturbed individual to obtain a new individual population. Then, a second position perturbation can be performed on the new individual population to prevent the population from falling into a local optimum after the position update. The process is shown in step S24d below.

[0173] S24d. Based on the fitness of each updated individual in the new individual population, perform a secondary position perturbation process on each updated individual to obtain the individual population at the (n+1)th iteration after the secondary position perturbation process.

[0174] In this embodiment, taking any updated individual as an example, the secondary position perturbation process can be described, but is not limited to the steps S24d1 to S24d4 below.

[0175] S24d1. For any updated individual in the new individual population, the perturbation intensity is calculated based on the fitness of that updated individual; in specific implementation, the perturbation intensity of that updated individual is: In the formula, For the disturbance intensity, For the fitness of any updated individual, This represents the perturbation coefficient, with values ​​between [0,1].

[0176] After calculating the perturbation intensity of any of the updated individuals, a second position random number can be generated, as shown in step S24d2 below.

[0177] S24d2. Generate a second position random number; in this embodiment, the value range of the second position random number is [0,1]; then, the second position random number can be compared with the perturbation probability so that different perturbation methods can be adopted according to the comparison result, the process is as shown in the following steps S24d3~S24d4.

[0178] S24d3. Determine whether the second position random number is less than the perturbation probability; in this embodiment, the perturbation probability is a preset value, which can be specifically set according to actual use; wherein, the two position perturbation methods are as shown in step S24d4 below.

[0179] S24d4. If so, then from the new individual population, determine the updated individual corresponding to the minimum fitness as the designated individual, and perform secondary position perturbation processing on any updated individual according to the second position random number, the perturbation intensity and the designated individual to obtain the secondary perturbation individual corresponding to any updated individual.

[0180] In this embodiment, for example, but not limited to, the following formula (12) can be used to perform secondary position perturbation processing on any of the updated individuals.

[0181] (12)

[0182] In equation (12), This represents the secondary perturbation individual corresponding to any of the updated individuals. Represents any updated individual, This represents a random number in the second position. Indicates a specific individual.

[0183] Similarly, when the second position random number is greater than or equal to the perturbation probability, two updated individuals are randomly selected from the new individual population as the first random individual and the second random individual. Based on the first random individual, the second random individual, the second position random number, and the perturbation intensity, a secondary position perturbation process is performed on any updated individual to obtain the secondary perturbation individual corresponding to any updated individual.

[0184] Optionally, for example, but not limited to, the following formula (13) can be used to perform secondary position perturbation processing on any of the updated individuals.

[0185] (13)

[0186] In equation (13), These represent the first random individual and the second random individual, respectively.

[0187] Thus, through the aforementioned steps S24d1 to S24d4, the secondary position perturbation of the new individual population can be completed, resulting in several perturbed individuals. Then, these perturbed individuals can be used to form the individual population for the (n+1)th iteration. After that, the aforementioned steps S22 to S24 can be repeated until the individual population for the nth iteration is in a globally converged state, at which point the search method is switched, as shown in step S25 below.

[0188] S25. Increment n by 1 and reacquire the individual population at the nth iteration until the individual population at the nth iteration is in a globally converged state, thus obtaining the globally optimal population; In this embodiment, when the global search converges, it is necessary to switch to the local search mode, that is, to perform a local fine search within the interval where the global optimal value is located, so as to quickly find the globally optimal individual within the interval, as shown in step S26 below.

[0189] S26. Using the minimum envelope entropy of signal decomposition as the search objective, perform iterative local search processing on the globally optimal population until the iteration meets the stopping condition, and obtain the globally optimal individual; in this embodiment, the whale optimization algorithm with high local search accuracy is used as an example to perform iterative local search processing on the globally optimal population; the process is as shown in the following steps S26a to S26j.

[0190] S26a. Obtain the search population at the k-th local search, where when k is 1, the search population at the k-th local search is the global optimal population.

[0191] S26b. Calculate the fitness of each individual in the search population during the k-th local search and determine the global optimal fitness during the k-th local search. In this embodiment, the fitness calculation process for each individual can be referred to any of the initial individuals mentioned above, and will not be repeated here. After obtaining the fitness of each individual, the global optimal fitness during the k-th local search can be determined. That is, first, the minimum fitness during the k-th local search is selected, and then it is determined whether the minimum fitness is less than the global optimal fitness during the (k-1)-th local search. If so, the minimum fitness during the k-th local search is taken as the global optimal fitness during the k-th local search; otherwise, the global optimal fitness during the (k-1)-th local search is taken as the global optimal fitness during the k-th local search. Of course, the global optimal fitness during the first local search is the minimum fitness during the first local search.

[0192] After obtaining the global optimal fitness at the current iteration, it can be determined whether the stopping condition of the local search is met, as shown in step S26c below.

[0193] S26c. Determine whether the stopping condition for the local search is met; in this embodiment, the stopping condition is that the global optimal fitness at the k-th local search is less than the fitness threshold, or k reaches the maximum number of local searches; if the stopping condition is not met, mutation operation and position update are required, and the process is as shown in steps S26d to S26i below.

[0194] S26d. If not, calculate the fitness variance of the search population during the k-th local search. In this embodiment, the fitness variance is used to determine the diversity of the local search. If the fitness variance is less than the variance threshold, it may be trapped in a local optimum within a certain search segment of the global optimum and cannot find the optimum value in the entire interval. In this case, population mutation is required, and the process is shown in steps S26e and S26f below.

[0195] S26e. Determine whether the fitness variance is less than the variance threshold.

[0196] S26f. If so, then a mutation operation is performed on the search population during the k-th local search to obtain a mutated population. In this embodiment, when the fitness variance is greater than or equal to the variance threshold, the following step S26g is directly performed; otherwise, a mutation operation is required. Taking any search individual as an example, the mutation formula is as follows:

[0197] (14)

[0198] In equation (14), This represents the variant individual corresponding to any searched individual. This refers to any of the searched individuals. This represents a random number that follows a Gaussian distribution.

[0199] After completing the mutation operation of each search individual based on the aforementioned formula (14), the convergence factor and update weight for the current local search can be determined, as shown in step S26g below.

[0200] S26g. Calculate the convergence factor and update the weights at the k-th local search. In practical applications, this embodiment provides a piecewise nonlinear convergence factor so that the convergence factor rapidly decreases to a small value with the number of local searches, thereby maintaining strong local exploitation capability within the interval where the global optimum converges. The formula for calculating the convergence factor is as follows:

[0201] (15)

[0202] In equation (15), This represents the convergence factor during the k-th local search. Indicates the maximum number of local searches. This is a non-linear adjustment coefficient, with a value of 2.

[0203] Similarly, for example, but not limited to, the following formula (16) can be used to calculate the update weight during the k-th local search.

[0204] (16)

[0205] In equation (16), denoted as the update weight during the k-th local search; it can be seen from formula (16) that the weight increases with the number of local searches, indicating that the prey selected through iterations is getting closer and closer to the optimal value, that is, the optimal solution of the current population, and the attraction to whales in the population is getting stronger and stronger, which can enable individual whales to find prey more accurately, thereby improving the speed and accuracy of local search.

[0206] After calculating the convergence factor at the kth local search, the coefficient vector of each variant individual can be calculated, as shown in step S26h below.

[0207] S26h. Based on the convergence factor, calculate the coefficient vector of each mutated individual during the k-th local search; in this embodiment, the formula for calculating the coefficient vector is the same as that of the traditional whale algorithm, and the coefficient vector is... The calculation process will not be elaborated here.

[0208] After obtaining the coefficient vectors of each mutant individual, the position of each mutant individual can be updated, as shown in step S26i below.

[0209] S26i. Based on the updated weights and the coefficient vectors of each mutant individual, the position of each mutant individual is updated to obtain several updated individuals, which are then used to form the search population for the (k+1)th local search.

[0210] In practice, examples, including but not limited to the following formula, can be used to update the position of any mutated individual.

[0211] (17)

[0212] In equation (17), This represents the updated individual corresponding to any mutated individual. This represents the mutant with the lowest fitness among all mutants during the k-th local search. This represents a coefficient vector corresponding to any of the mutated individuals. Indicates intermediate parameters. , Represents another coefficient vector, Represents a random number between (0, 1). express and The distance vector between them This represents the constant coefficient that defines the shape of the logarithmic spiral. Represents a random number between [-1, 1]. It represents any variant individual that is different from any other variant individual.

[0213] Thus, by using the aforementioned formula (17) to update the position of each mutated individual, the search population at the (k+1)th local search can be obtained. Then, the aforementioned steps S26a to S26i can be repeated until the local search stopping condition is met, and the globally optimal individual is obtained. The process is shown in step S26j below.

[0214] S26j. Increment k by 1 and reacquire the search population at the time of the kth local search until the stopping condition of the local search is met. Then, take the search individual corresponding to the global optimal fitness at the time the stopping condition is met as the global optimal individual.

[0215] After obtaining the globally optimal individual based on the aforementioned steps, the optimal decomposition parameters can be determined based on the globally optimal individual, as shown in step S27 below.

[0216] S27. Determine the optimal decomposition parameters based on the globally optimal individual; in this embodiment, the optimal decomposition parameters are determined based on the position vector of the globally optimal individual.

[0217] Therefore, through the aforementioned steps S21 to S27 and their steps, the optimal decomposition parameters of the generator set operation vibration signal can be obtained by optimizing the parameters of the ensemble empirical mode decomposition with the minimum envelope entropy as the optimization objective. Then, the signal decomposition of the operation vibration signal can be performed based on the optimal decomposition parameters, as shown in step S3 below.

[0218] S3. Based on the optimal decomposition parameters, the running vibration signal is subjected to ensemble empirical mode decomposition to obtain several component signals; in this embodiment, ensemble empirical mode decomposition is a common method for signal denoising, and its decomposition process will not be described in detail.

[0219] After obtaining several component signals, effective components can be filtered. In this embodiment, signal filtering is performed by calculating the signal distribution trend characteristics between the component signals and the original signal. Optionally, the signal filtering process can be, but is not limited to, the steps shown in step S4 below.

[0220] S4. Calculate the signal distribution trend characteristics between each component signal and the running vibration signal, and determine the effective component from several component signals based on the signal distribution trend characteristics between each component signal and the running vibration signal; In specific implementation, this embodiment takes any component signal as an example to illustrate the calculation process of the signal distribution trend characteristics, and the process may be, but is not limited to, the following steps S41 to S44.

[0221] S41. For any component signal, calculate the first mean of the sorting indices of all signal values ​​in the component signal and the second mean of the sorting indices of all signal values ​​in the running vibration signal, wherein the sorting index of any signal value in the component signal is the sorting sequence number of the signal value in the component signal; in this embodiment, assuming any component signal is [3,2,4], then the sorting index of signal value 3 is 1, the sorting index of signal value 2 is 2, and the sorting index of signal value 4 is 3. Based on this, the first mean of the sorting index of the component signal is: (1+2+3) / 3=2; of course, the calculation process of the second mean of the sorting index of the running vibration signal is also the same, and will not be repeated here.

[0222] After obtaining the mean of the sorting index of the signal values ​​in any component signal and the running vibration signal, the difference of the sorting index can be calculated, as shown in step S42 below.

[0223] S42. Calculate the difference between the sorting index of each signal value in any component signal and the first mean to obtain several first sorting index differences, and calculate the difference between the sorting index of each signal value in the running vibration signal and the second mean to obtain several second sorting index differences.

[0224] After calculating several first sorting index differences and several second sorting index differences, the distribution difference degree and distribution dispersion degree can be calculated based on these, as shown in step S43 below.

[0225] S43. Based on several first sorting index differences and several second sorting index differences, calculate the distribution difference degree and distribution dispersion degree between any component signal and the running vibration signal respectively; in this embodiment, for the i-th first sorting index difference, multiply the i-th first sorting index difference by the i-th second sorting index difference to obtain the i-th index distribution difference (i.e., rank difference); then, increment i by 1, and multiply the i-th first sorting index difference by the i-th second sorting index difference again, until i equals B, to obtain B index distribution differences. Finally, sum the B index distribution differences to obtain the distribution difference degree between any component signal and the running vibration signal (where B is the number of first sorting index differences); thus, if the rank distributions are similar, the value of this term is smaller, and conversely, if the rank distribution differences are large, the value of this term is larger.

[0226] To determine the degree of dispersion, the square of the difference between each first sorting index is first calculated. Then, the squares of the differences between each first sorting index are summed to obtain the first sum, and the square root of the first sum is taken to obtain the first square root value. Similarly, the squares of the differences between each second sorting index are summed to obtain the second sum. Next, the square root of the second sum is taken to obtain the second square root value. Finally, the product of the first square root value and the second square root value is taken as the degree of dispersion.

[0227] In this embodiment, if the rank distributions of the two are concentrated, the value of this term is smaller; conversely, if the rank distributions are dispersed, the value of this term is larger. Thus, the signal distribution trend characteristics can be calculated by calculating the distribution difference and dispersion, as shown in step S44 below.

[0228] S44. Calculate the signal distribution trend characteristics based on the distribution difference and distribution dispersion. In this embodiment, the ratio between the distribution difference and distribution dispersion is used as the signal distribution trend characteristics. The distribution trend characteristics of two time series are measured by comparing the degree of difference and dispersion of their rank distributions. If the rank distributions are similar and concentrated, the distribution trend characteristics are small, and the similarity between the two is high. If the rank distributions are significantly different and dispersed, the distribution trend characteristics are large, and the similarity between the two is low.

[0229] Thus, after obtaining the signal distribution trend characteristics between any component signal and the running vibration signal, the effective components can be screened based on this, as shown in steps S45 to S48 below.

[0230] S45. For any component signal, determine the fluctuation trend characteristics between the any component signal and the operating vibration signal; in specific implementation, for example, but not limited to, the following formula can be used to calculate the fluctuation trend characteristics.

[0231] (18)

[0232] In equation (18), Indicates the characteristics of fluctuation trends. This represents the t-th value in the signal obtained after taking the first derivative of the vibration signal. Let T represent the t-th value in the signal obtained after taking the first derivative of any component signal, where T represents the signal length of the running vibration signal.

[0233] After calculating the fluctuation trend characteristics, the amplitude difference can be calculated, as shown in step S46 below.

[0234] S46. Calculate the amplitude difference between any component signal and the running vibration signal; in this embodiment, the amplitude difference can be calculated by, but is not limited to, the following formula (19).

[0235] (19)

[0236] In equation (19), Indicates the degree of amplitude difference. This represents the t-th signal value in the vibration signal. This represents the t-th signal value in any component signal.

[0237] After calculating the amplitude difference between any component signal and the running vibration signal, the similarity between the two can be calculated by combining the aforementioned signal distribution trend characteristics and fluctuation trend characteristics, as shown in step S47 below.

[0238] S47. Based on the fluctuation trend characteristics, the amplitude difference degree, and the signal distribution trend characteristics, calculate the correlation between any component signal and the operating vibration signal, and obtain the correlation between each component signal and the operating vibration signal after polling all component signals.

[0239] In this embodiment, for example, but not limited to, the following formula (20) can be used to calculate the correlation between any component signal and the running vibration signal.

[0240] (20)

[0241] In equation (20), This indicates the correlation between any component signal and the operating vibration signal. This indicates the trend characteristics of signal distribution; among which, The smaller the value, the higher the correlation between any component signal and the operating vibration signal, i.e., the higher the similarity.

[0242] Thus, after calculating the correlation between each component signal and the running vibration signal using the aforementioned formula (20), the effective components can be screened based on the correlation, as shown in step S48 below.

[0243] S48. Based on the correlation between each component signal and the running vibration signal, the effective component is determined from several component signals; in this embodiment, the smallest correlation is selected, and then the component signal corresponding to the smallest correlation is taken as the truncated component; finally, the truncated component, as well as the component signal before the truncated component among several component signals, are taken as the effective component.

[0244] Therefore, through the aforementioned steps S41 to S48, effective components can be selected from several component signals, thereby removing invalid components. Based on this, the effective components are reconstructed to obtain the reconstructed vibration signal, as shown in step S5 below.

[0245] S5. The effective components are reconstructed to obtain a reconstructed vibration signal, and the reconstructed vibration signal is processed by feature extraction to obtain the fault characteristics of the gas turbine generator set. In this embodiment, after the effective components are reconstructed to obtain the reconstructed vibration signal, the noise component in the original vibration signal can be removed. Then, fault features can be extracted so that the fault identification of the gas turbine generator set can be performed based on the fault features.

[0246] In practical applications, examples include, but are not limited to, performing Hilbert envelope demodulation on the reconstructed vibration signal to extract the frequency and its harmonics of the vibration signal, thereby forming fault features. Of course, the aforementioned feature extraction is a common method for extracting fault features of vibration signals, and its process will not be elaborated on one by one.

[0247] After obtaining the fault characteristics, they can be input into the pre-trained fault monitoring model to obtain the operation fault monitoring results of the gas turbine generator set. The process is shown in step S6 below.

[0248] S6. Input the fault features into the fault monitoring model to obtain the operational fault monitoring results of the gas turbine generator set. In this embodiment, for example, but not limited to, a trained SVM model can be used as the fault monitoring model. Thus, by inputting the fault features of the operating vibration signals of different parts of the gas turbine generator set into the trained SVM model, fault identification results of different parts can be obtained. Finally, the fault identification results of different parts can be used to form the operational fault monitoring results. In this way, based on the operational fault monitoring results, the operation management of the gas turbine generator set can be carried out, so that when a fault is identified in a certain part, the early warning and maintenance process can be initiated in a timely manner.

[0249] Therefore, through the online monitoring method for gas turbine generator set faults described in detail in steps S1 to S5 above, the present invention can improve the noise reduction effect of generator set vibration signals, thereby ensuring the accuracy of generator set fault identification. Therefore, it is very suitable for large-scale application and promotion.

[0250] like Figure 2 As shown, the second aspect of this embodiment provides a hardware system for implementing the online fault monitoring method for gas turbine generator sets described in the first aspect of the embodiment, comprising:

[0251] The acquisition unit is used to acquire the operating vibration signals of the gas turbine generator set.

[0252] The parameter optimization unit is used to perform ensemble empirical mode decomposition parameter optimization processing on the running vibration signal with the goal of minimizing the envelope entropy of signal decomposition, so as to obtain the optimal decomposition parameters of the running vibration signal.

[0253] The signal decomposition unit is used to perform ensemble empirical mode decomposition on the running vibration signal based on the optimal decomposition parameters to obtain several component signals.

[0254] The signal reconstruction unit is used to calculate the signal distribution trend characteristics between each component signal and the operating vibration signal, and to determine the effective component from several component signals based on the signal distribution trend characteristics between each component signal and the operating vibration signal.

[0255] The signal reconstruction unit is also used to reconstruct the effective components to obtain a reconstructed vibration signal, and to perform feature extraction processing on the reconstructed vibration signal to obtain the fault characteristics of the gas turbine generator set.

[0256] The fault monitoring unit is used to input the fault characteristics into the fault monitoring model to obtain the operational fault monitoring results of the gas turbine generator set.

[0257] The working process, working details and technical effects of the system provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0258] like Figure 3 As shown, the third aspect of this embodiment provides an online fault monitoring device for a gas turbine generator set. Taking the device as an electronic device as an example, it includes: a memory, a processor, and a transceiver connected in sequence. The memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the online fault monitoring method for a gas turbine generator set as described in the first aspect of the embodiment.

[0259] For specific examples, the memory may include, but is not limited to, random access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented using at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor, also known as the CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state.

[0260] In some embodiments, the processor may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. For example, the processor may not be limited to microprocessors of the STM32F105 series, reduced instruction set computer (RISC) microprocessors, x86 architecture processors, or processors with integrated neural network processing units (NPUs). The transceiver may be, but is not limited to, a Wi-Fi transceiver, a Bluetooth transceiver, a General Packet Radio Service (GPRS) transceiver, a ZigBee (a low-power LAN protocol based on the IEEE 802.15.4 standard) transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver. Furthermore, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0261] The working process, working details and technical effects of the electronic device provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0262] The fourth aspect of this embodiment provides a storage medium for storing instructions containing the online fault monitoring method for gas turbine generator sets as described in the first aspect of the embodiment. That is, the storage medium stores instructions, and when the instructions are run on a computer, the online fault monitoring method for gas turbine generator sets as described in the first aspect of the embodiment is executed.

[0263] The storage medium refers to a carrier for storing data, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or memory sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0264] The working process, working details and technical effects of the storage medium provided in this embodiment can be found in the first aspect of the embodiment, and will not be repeated here.

[0265] The fifth aspect of this embodiment provides a computer program product containing instructions that, when executed on a computer, cause the computer to perform the online fault monitoring method for gas turbine generator sets as described in the first aspect of this embodiment, wherein the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device.

[0266] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of on-line monitoring of a gas turbine generator set failure, characterized by, The method comprises the following steps: acquiring an operating vibration signal of a gas turbine generator set; performing parameter optimization processing on the operating vibration signal by ensemble empirical mode decomposition (EEMD) to obtain optimal decomposition parameters of the operating vibration signal, including: initializing the decomposition parameters of the EEMD to obtain multiple sets of decomposition parameters, and constructing an optimal initial population according to the multiple sets of decomposition parameters, wherein each individual in the optimal initial population corresponds to a set of decomposition parameters; acquiring an individual population at the nth iteration, and calculating the fitness of each individual in the individual population at the nth iteration, wherein the fitness of any individual is the minimum envelope entropy of a plurality of component signals obtained by performing EEMD on the operating vibration signal based on the decomposition parameters corresponding to the individual, wherein the initial value of n is 1, and when n is 1, the individual population at the nth iteration is the optimal initial population; determining whether the individual population at the nth iteration is in a global convergence state according to the fitness of each individual; if not, performing a global search on the individual population at the nth iteration to obtain an individual population at the (n+1)th iteration, including: performing a position disturbance processing on each individual in the individual population at the nth iteration to obtain a disturbed population, wherein the disturbed population is obtained by: for any individual in the individual population at the nth iteration, selecting a target individual from the individual population at the nth iteration, wherein the target individual and the individual are different from each other; calculating a first distance between the individual and a best individual, and a second distance between the target individual and the best individual, wherein the best individual is the individual with the minimum fitness in the individual population at the nth iteration; calculating a first disturbance control factor according to the first distance and the second distance; obtaining a global search range of the individual, and calculating a first disturbance intermediate parameter according to the global search range, the first distance and the individual, and calculating a second disturbance intermediate parameter according to the global search range, the second distance and the target individual; calculating a second disturbance control factor according to the first disturbance intermediate parameter and the second disturbance intermediate parameter; performing a position disturbance processing on the individual based on the first disturbance control factor, the second disturbance control factor and the global search range to obtain a disturbed individual corresponding to the individual, and using the disturbed individual corresponding to each individual to form the disturbed population after all individuals in the individual population at the nth iteration are polled; calculating a global search evolution value of the disturbed population according to the current iteration number and a maximum iteration number of the global search; performing a global search processing on the disturbed population based on the global search evolution value to obtain a new individual population; performing a secondary position disturbance processing on each updated individual in the new individual population according to the fitness of each updated individual to obtain the individual population at the (n+1)th iteration after the secondary position disturbance processing. increment n by 1, and reacquire the individual population at the n th iteration until the individual population at the n th iteration is in a global convergence state, to obtain a global optimal population; taking the signal decomposition minimum envelope entropy as a search target, performing an iterative local search process on the global optimal population until the iteration meets a stop condition, to obtain a global optimal individual; determining the optimal decomposition parameter according to the global optimal individual; performing an ensemble empirical mode decomposition process on the running vibration signal based on the optimal decomposition parameter, to obtain a plurality of component signals; calculating a signal distribution trend feature between each component signal and the running vibration signal, and determining an effective component from the plurality of component signals according to the signal distribution trend feature between each component signal and the running vibration signal; performing signal reconstruction on the effective component to obtain a reconstructed vibration signal, and performing feature extraction processing on the reconstructed vibration signal to obtain a fault feature of the gas turbine generator set; inputting the fault feature into a fault monitoring model to obtain an operation fault monitoring result of the gas turbine generator set.

2. The method of claim 1, wherein, According to the current iteration number and the maximum iteration number of global search, the global search evolution value of the perturbed population is calculated, including: calculating the ratio between the current iteration number and the maximum iteration number, and calculating the difference between the ratio threshold and the ratio to obtain the evolution coefficient; obtaining a first evolution random number, and determining the global search evolution value according to the evolution coefficient and the first evolution random number; Correspondingly, based on the global search evolution value, the perturbed population is processed once to obtain a new individual population, including: determining whether the global search evolution value is greater than a first evolution threshold; if not, determining whether the global search evolution value is greater than or equal to a second evolution threshold, wherein the second evolution threshold is less than the first evolution threshold; if yes, generating a second evolution random number; determining whether the second evolution random number is less than the second evolution threshold; if not, generating a global search coefficient, and selecting the best perturbed individual from the perturbed population, to perform a global search process on each perturbed individual in the perturbed population according to the best perturbed individual, the global search coefficient and the global search evolution value, to obtain an updated individual corresponding to each perturbed individual, so as to use each updated individual to form a new individual population.

3. The method of claim 2, wherein, If the global search evolution value is greater than the first evolution threshold, the method further includes: generating a plurality of first position random numbers for each perturbed individual in the perturbed population; calculating the average position of the individuals in the perturbed population; performing a global search process on each perturbed individual according to the plurality of first position random numbers, the average position of the individuals and the best perturbed individual, to obtain an updated individual corresponding to each perturbed individual; wherein, if the global search evolution value is less than the second evolution threshold, the method further includes: generating a third evolution random number, and determining whether the third evolution random number is greater than or equal to the second evolution threshold; If yes, a global search process is performed on each perturbation individual according to the best perturbation individual and the global search evolution value, to obtain an updated individual corresponding to each perturbation individual; otherwise, a global search process is performed on each perturbation individual according to the individual average position, the best perturbation individual, the global search coefficient and the global search evolution value, to obtain an updated individual corresponding to each perturbation individual.

4. The method of claim 1, wherein, According to the fitness of each updated individual in the new individual population, a secondary position perturbation process is performed on each updated individual, including: For any updated individual in the new individual population, a perturbation intensity is calculated according to the fitness of the any updated individual; A second position random number is generated; It is judged whether the second position random number is less than a perturbation probability; If yes, an updated individual corresponding to the minimum fitness is determined from the new individual population as a specified individual, and a secondary position perturbation process is performed on the any updated individual according to the second position random number, the perturbation intensity and the specified individual, to obtain a secondary perturbation individual corresponding to the any updated individual; Otherwise, two updated individuals are randomly selected from the new individual population as a first random individual and a second random individual, and a secondary position perturbation process is performed on the any updated individual according to the first random individual, the second random individual, the second position random number and the perturbation intensity, to obtain a secondary perturbation individual corresponding to the any updated individual.

5. The method of claim 1, wherein, A signal distribution trend feature between each component signal and the running vibration signal is calculated, including: For any component signal, a first mean value of ordering indexes of all signal values in the any component signal and a second mean value of ordering indexes of all signal values in the running vibration signal are calculated, wherein the ordering index of any signal value in the any component signal is the ordering serial number of the any signal value in the any component signal; A difference value between the ordering index of each signal value in the any component signal and the first mean value is calculated to obtain a plurality of first ordering index differences, and a difference value between the ordering index of each signal value in the running vibration signal and the second mean value is calculated to obtain a plurality of second ordering index differences; Based on the plurality of first ordering index differences and the plurality of second ordering index differences, a distribution difference degree and a distribution dispersion degree between the any component signal and the running vibration signal are respectively calculated; The signal distribution trend feature is calculated according to the distribution difference degree and the distribution dispersion degree.

6. The method of claim 1, wherein, An effective component is determined from a plurality of component signals according to the signal distribution trend feature between each component signal and the running vibration signal, including: For any component signal, a fluctuation trend feature between the any component signal and the running vibration signal is determined; An amplitude difference degree between the any component signal and the running vibration signal is calculated; According to the fluctuation trend feature, the amplitude difference degree and the signal distribution trend feature, the correlation between each component signal and the operation vibration signal is calculated, and after all component signals are polled, the correlation between each component signal and the operation vibration signal is obtained; According to the correlation between each component signal and the operation vibration signal, an effective component is determined from the several component signals.

7. An online monitoring system for gas turbine generator set failures, characterized by, The method for monitoring the fault of the gas turbine generator set in real time according to any one of claims 1-6, wherein the system comprises: an acquisition unit configured to acquire an operation vibration signal of the gas turbine generator set; a parameter optimization unit configured to perform parameter optimization processing on the operation vibration signal by using ensemble empirical mode decomposition (EEMD) to obtain an optimal decomposition parameter of the operation vibration signal, with the minimum envelope entropy of signal decomposition as an optimization objective; a signal decomposition unit configured to perform EEMD processing on the operation vibration signal based on the optimal decomposition parameter to obtain several component signals; a signal reconstruction unit configured to calculate a signal distribution trend feature between each component signal and the operation vibration signal, and determine an effective component from the several component signals according to the signal distribution trend feature between each component signal and the operation vibration signal; the signal reconstruction unit is further configured to perform signal reconstruction on the effective component to obtain a reconstructed vibration signal, and perform feature extraction processing on the reconstructed vibration signal to obtain a fault feature of the gas turbine generator set; a fault monitoring unit configured to input the fault feature into a fault monitoring model to obtain an operation fault monitoring result of the gas turbine generator set.

Images