Power module abnormality diagnosis method and system

Abstract

The application discloses a power module abnormality diagnosis method and system, and the application carries out optimization processing on the adjustment parameter in the initial wavelet threshold function by constructing a signal denoising evaluation function, so that the optimal adjustment parameter is obtained objectively and globally, which solves the problem that the denoising result is different for different people due to the experience difference of different operators, and ensures the objectivity and stability of the signal preprocessing stage; meanwhile, on the basis of the optimal wavelet threshold function denoising, the application further introduces a mixed form filtering algorithm to process the initial denoising signal, and the combined strategy can effectively filter out the pulse noise in the electromagnetic environment, and realizes deep suppression of the complex mixed noise; thus, the application significantly improves the denoising quality and robustness of the power signal through the objective parameter optimization and the mixed filtering strategy, so that the extracted power fault feature is more accurate, and finally the accuracy of the power module abnormality diagnosis is greatly improved.

Description

Technical Field

[0001] This invention belongs to the field of power supply fault detection technology, specifically relating to a method and system for diagnosing power module anomalies. Background Technology

[0002] As the core component of electrical equipment, power modules are indispensable in daily life. During use, various factors inevitably lead to various faults in power modules. Therefore, abnormal diagnosis of power modules and timely detection of faults are of great significance to ensuring the stable operation of power modules.

[0003] Currently, fault diagnosis requires acquiring power signals from the power module. However, in practical applications, these signals are inevitably affected by electromagnetic noise in the environment. Therefore, to improve the accuracy of power fault diagnosis, denoising of the acquired power signals is usually necessary before diagnosis. Currently, soft-threshold wavelet functions with adjustable parameters are commonly used for signal denoising, but this method has the following shortcomings: While traditional soft-threshold wavelet functions overcome the discontinuity problem of hard-threshold functions, making the processing of wavelet coefficients smoother, the adjustment parameters in soft-threshold wavelet functions are usually estimated with human experience. The value of this adjustment parameter has a significant impact on the signal denoising effect. Therefore, traditional techniques suffer from strong subjectivity and a lack of objective standards; different experienced individuals may provide completely different values, leading to inconsistent denoising results. Furthermore, it is impossible to find a globally optimal value for the adjustment parameter, resulting in poor denoising performance. Simultaneously, traditional wavelet denoising primarily targets white noise in electromagnetic noise, with poor filtering effect on other impulse noise, further reducing the denoising effect and affecting the accuracy of subsequent fault diagnosis. Therefore, given these shortcomings, providing a power module anomaly diagnosis method with high fault diagnosis accuracy has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a power module anomaly diagnosis method and system to solve the problem of poor noise reduction effect in the prior art, which affects the accuracy of power fault diagnosis.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: Firstly, a method for diagnosing power module anomalies is provided, including: Obtain the power signal from the power module; An initial wavelet threshold function for the power supply signal and a signal denoising evaluation function based on the time-frequency domain are constructed. Based on the signal denoising evaluation function, the adjustment parameters in the initial wavelet threshold function are optimized to obtain the optimal adjustment parameters. Then, the initial wavelet threshold function is optimized using the optimal adjustment parameters to obtain the optimal wavelet threshold function. The power signal is denoised using the optimal wavelet threshold function to obtain an initial denoised signal. A hybrid morphological filtering algorithm is used to perform morphological filtering on the initial denoised signal to obtain the denoised signal. The denoised signal is subjected to fault feature extraction processing to obtain power supply fault features; The power failure characteristics are input into the power failure detection model to obtain the abnormal diagnosis results of the power module.

[0006] Based on the above-disclosed content, this invention constructs a signal denoising evaluation function to optimize the adjustment parameters in the initial wavelet threshold function, thereby obtaining objective and globally optimal adjustment parameters. This improvement completely eliminates the subjectivity and uncertainty of traditional techniques that rely on human experience to set initial values, and solves the problem of varying denoising results due to differences in the experience of different operators. This ensures the objectivity and stability of the signal preprocessing stage, laying a solid foundation for the accuracy of subsequent fault diagnosis. Furthermore, based on the optimal wavelet threshold function denoising, this invention further introduces a hybrid morphological filtering algorithm to process the initial denoised signal. This combined strategy effectively overcomes the limitation of traditional wavelet denoising, which is only good at handling white noise, and can effectively filter out impulse noise in the electromagnetic environment, achieving deep suppression of complex mixed noise. Therefore, through objective parameter optimization and a hybrid filtering strategy, this invention significantly improves the denoising quality and robustness of power signals, resulting in more accurate extracted power fault features and ultimately greatly improving the accuracy of power module anomaly diagnosis.

[0007] In one possible design, a signal denoising evaluation function is constructed, including: Establish the residual sequence between the power signal and the denoised power signal; The residual sequence is subjected to multiple lag processes to obtain several lag residual sequences; Calculate the autocorrelation coefficient of each lagged residual sequence, and establish the lagged correlation function of the residual sequence based on each autocorrelation coefficient; Establish the spectral flatness function of the residual sequence; The signal denoising evaluation function is constructed using the hysteresis correlation function and the spectral flatness function.

[0008] In one possible design, based on the signal denoising evaluation function, the adjustment parameters in the initial wavelet threshold function are optimized to obtain the optimal adjustment parameters, including: Obtain the individual population at the t-th iteration, where when t is 1, the individual population at the t-th iteration is the initial population. Each initial individual in the initial population is configured with a position vector and a velocity vector, and the position vector corresponding to any initial individual is used to represent a set of initial adjustment parameters. Based on each individual in the population at the t-th iteration, a wavelet threshold function corresponding to each individual is constructed, and the power signal is denoised using each wavelet threshold function to obtain the denoised power signal corresponding to each individual. The fitness of each individual is calculated based on the signal denoising evaluation function and using the power supply signal and the denoised power supply signal corresponding to each individual. Based on the fitness of each individual, the globally optimal individual at the t-th iteration is determined; Determine if the iteration stopping condition is met; If not, a dual-modal update mechanism is used to update the velocity vector of each individual at the t-th iteration, thus obtaining the updated velocity vector of each individual. The updated velocity vectors of each individual are used to update the position vectors of each individual to obtain the population of individuals at the (t+1)th iteration. Increment t by 1 and reacquire the individual population at the t-th iteration until the iteration stopping condition is met. Based on the globally optimal individual at the iteration stopping condition, determine the optimal adjustment parameter.

[0009] In one possible design, a bimodal update mechanism is used to update the velocity vectors of each individual at the t-th iteration, including: Calculate the search weight and search control factor at the t-th iteration; For any individual in the population at the t-th iteration, the search probability of generating that individual; Determine if the search probability is less than the probability threshold; If so, the velocity vector of any individual is updated based on the search weight and search control factor, using a perturbation global update mechanism, to obtain the updated velocity vector corresponding to that individual. Otherwise, the velocity vector of any individual is updated based on the search weight and search control factor, using a local update mechanism, to obtain the updated velocity vector corresponding to that individual. This process is repeated until all individuals at the t-th iteration have been polled, at which point the updated velocity vectors for each individual are obtained.

[0010] In one possible design, based on search weights and search control factors, and employing a perturbation-based global update mechanism, the velocity vector of any individual is updated, including: Update the velocity vector of any individual according to the following formula; ; In the formula, This represents the velocity vector of any of the individuals. Indicates the updated , This represents the search weight at the t-th iteration. Both represent the search control factors at the t-th iteration. This represents the historical best position vector of any given individual. This represents the position vector of any of the individuals. Let represent the position vector corresponding to the globally optimal individual at the t-th iteration. Scaling factor , A random number between [0, 1]; Accordingly, based on search weights and search control factors, and employing a local update mechanism, the velocity vector of any individual is updated, including: Update the velocity vector of any individual according to the following formula; ; In the formula, The oscillation coefficient is, and .

[0011] In one possible design, a hybrid morphological filtering algorithm is used to perform morphological filtering on the initial denoised signal to obtain a denoised signal, including: Obtain the sequence of structure elements; Based on the structuring element sequence, the initial denoised signal is sequentially subjected to opening and closing operations to obtain the first filtered signal; Based on the structuring element sequence, the initial denoised signal is sequentially subjected to closing and opening operations to obtain the second filtered signal; The first filtered signal and the second filtered signal are averaged to obtain the denoised signal.

[0012] In one possible design, fault feature extraction processing is performed on the denoised signal to obtain power supply fault features, including: The optimal wavelet decomposition level is obtained, and the denoised signal is processed by wavelet decomposition using the optimal wavelet decomposition level to obtain wavelet signals at different decomposition scales. From the wavelet signals at all decomposition scales, extract the wavelet signal at the largest decomposition scale, and obtain all frequency band sequences in the extracted wavelet signal; For any frequency band sequence, the sequence is divided into several sub-frequency bands. Based on each sub-band, the distribution dispersion of each sub-band is calculated; Based on the distribution dispersion of each sub-band, the sub-band used for power supply fault diagnosis is extracted from several sub-bands and used as the key sub-band. The energy of each key sub-band is calculated, and after polling all frequency band sequences, the energy of all key sub-bands in each frequency band sequence is obtained. The frequency domain characteristics of the power signal are formed by utilizing the energy of all key sub-bands in each frequency band sequence. The power signal is processed by time-domain feature extraction to obtain the time-domain features of the power module; The power supply fault characteristics are formed by using the frequency domain characteristics and the time domain characteristics.

[0013] Secondly, a power module anomaly diagnosis system is provided, including: Acquisition unit, used to acquire power signals from the power module; The construction unit is used to construct the initial wavelet threshold function of the power signal and the signal denoising evaluation function based on the time and frequency domain; The parameter optimization unit is used to optimize the adjustment parameters in the initial wavelet threshold function based on the signal denoising evaluation function to obtain the optimal adjustment parameters, and to optimize the initial wavelet threshold function using the optimal adjustment parameters to obtain the optimal wavelet threshold function. The denoising unit is used to denoise the power signal using the optimal wavelet threshold function to obtain the initial denoised signal. The denoising unit is also used to perform morphological filtering on the initial denoising signal using a hybrid morphological filtering algorithm to obtain the denoised signal. The feature extraction unit is used to perform fault feature extraction processing on the denoised signal to obtain power fault features; An anomaly diagnosis unit is used to input the power failure characteristics into the power failure detection model to obtain the anomaly diagnosis results of the power module.

[0014] Thirdly, a power module anomaly diagnosis device 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 power module anomaly diagnosis method as described in the first aspect or any possible design of the first aspect.

[0015] Fourthly, a storage medium is provided, on which instructions are stored, which, when executed on a computer, perform the power module anomaly diagnosis method as described in the first aspect or any possible design of the first aspect.

[0016] Fifthly, a computer program product containing instructions is provided, which, when executed on a computer, cause the computer to perform the power module anomaly diagnosis method as described in the first aspect or any possible design of the first aspect.

[0017] Beneficial effects: (1) This invention constructs a signal denoising evaluation function and optimizes the adjustment parameters in the initial wavelet threshold function to obtain objective and global optimal adjustment parameters. This improvement completely eliminates the subjectivity and uncertainty of the traditional technology that relies on human experience to set the initial value, and solves the problem that the denoising results vary from person to person due to the difference in experience of different operators. It ensures the objectivity and stability of the signal preprocessing stage and lays a solid foundation for the accuracy of subsequent fault diagnosis. At the same time, based on the optimal wavelet threshold function denoising, this invention further introduces a hybrid morphological filtering algorithm to process the initial denoised signal. This combined strategy effectively overcomes the limitation of traditional wavelet denoising that is only good at processing white noise. It can effectively filter out impulse noise in the electromagnetic environment and achieve deep suppression of complex mixed noise. Thus, this invention significantly improves the denoising quality and robustness of the power signal through objective parameter optimization and hybrid filtering strategy, thereby making the extracted power fault features more accurate and ultimately greatly improving the accuracy of power module anomaly diagnosis. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating the steps of the power module anomaly diagnosis method provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the power module fault diagnosis system provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] 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.

[0020] 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.

[0021] 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.

[0022] Example: See Figure 1 As shown, the power module anomaly diagnosis method provided in this embodiment can be executed, but is not limited to, by a computer device with certain computing resources, such as a server, edge computer, personal computer (PC, which refers to a multi-purpose computer of a size, price and performance suitable for personal use; desktop computers, laptops to mini-laptops and tablets and ultrabooks are all personal computers), smartphone or personal digital assistant (PDA) and other electronic devices. It is understood that the aforementioned execution subject 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 S7 below.

[0023] S1. Acquire the power signal of the power module; in specific implementation, for example, but not limited to, the power signal of the power module can be acquired according to a preset sampling period. At the same time, the power signal can include, but not limited to, the voltage signal and current signal of the power module. After the power signal is acquired, the power signal can be denoised, that is, the white noise and impulse noise in the power signal can be removed. In this embodiment, a combination of improved wavelet denoising algorithm and hybrid morphological filtering algorithm is used to remove white noise and impulse noise, thereby achieving deep suppression of complex mixed noise in the power signal, and thus providing an accurate judgment basis for subsequent power anomaly diagnosis.

[0024] Optionally, in order to overcome the subjectivity and uncertainty caused by relying on manual experience to select the adjustment parameters in the wavelet threshold function when removing white noise from the power signal, this embodiment constructs a signal denoising evaluation function, and then optimizes the adjustment parameters based on the signal denoising evaluation function to obtain the adjustment parameters with the best denoising effect, thereby filtering out white noise in the power signal to the greatest extent.

[0025] The construction process of the initial wavelet threshold function and the signal denoising evaluation function is shown in step S2 below.

[0026] S2. Construct the initial wavelet threshold function for the power signal and the signal denoising evaluation function based on the time-frequency domain. In practical applications, although the traditional improved wavelet threshold function compresses larger wavelet coefficients, has good continuity and stability in denoising, and strong denoising capability, it is limited by the discontinuity of its first derivative. After operation, a constant deviation will appear between the original and improved wavelet coefficients. This deviation significantly affects the approximation between the original and reconstructed signals, and may eventually lead to signal edge distortion. At the same time, traditional wavelet function threshold denoising discards wavelet coefficients smaller than the threshold, which will also filter out useful signals mixed in with noise, causing a certain deviation between the reconstructed signal and the original signal. Therefore, to solve the above-mentioned shortcomings, this embodiment provides a new initial wavelet threshold function, as shown below.

[0027] ; In the formula, This represents the k-th wavelet coefficient at the i-th decomposition scale after wavelet decomposition of the power signal. Indicates the updated , The wavelet threshold at the i-th decomposition scale, All represent adjustment parameters, among which, ,and This represents the number of high-frequency coefficients at the i-th decomposition scale.

[0028] In this embodiment, the aforementioned three adjustment parameters are used to increase the noise reduction sensitivity of the function in actual use, wherein, All are greater than 0, representing the shape control parameters of the function, while The parameter between (0,1) is the approximation control parameter for this function, i.e., when... When the value is 1, the function approximates the traditional hard thresholding function, while when... When the value is 0, it is closer to the soft threshold parameter.

[0029] Furthermore, as can be seen from the aforementioned formula, the hyperbolic tangent function is used in the formula. This function is infinitely differentiable over the entire real number field, and its derivative is continuous, ensuring the smoothness of the entire threshold function. Furthermore, this formula... By using a continuous structural design (i.e., using tanh and polynomials), the discontinuity problem of the traditional hard threshold function at the threshold is avoided, and the oscillation phenomenon of the reconstructed signal is eliminated. Therefore, the entire wavelet threshold function is continuous and smooth.

[0030] Furthermore, since tanh is an analytic function and a polynomial... Waiting When the value is greater than 0, it is also smooth. Therefore, the entire threshold function has a high-order continuous derivative in the domain. This high-order differentiability makes the wavelet threshold function smoother when processing wavelet coefficients, avoiding signal distortion caused by abrupt changes in derivatives. In addition, three adjustment parameters are introduced into the entire threshold function, including two shape adjustment parameters and a parameter that controls the degree to which the function approximates the hard or soft threshold. Therefore, by adjusting these parameters, the degree of shrinkage of wavelet coefficients can be flexibly controlled to achieve adaptive shrinkage.

[0031] Therefore, the wavelet threshold function constructed in this embodiment has high-order differentiability and adaptive shrinkage capability of wavelet coefficients, retaining the noise reduction advantages of both hard and soft threshold functions. Furthermore, it can be adjusted... The size of the wavelet threshold function makes the wavelet threshold function in The system maintains continuity to prevent oscillations and avoids deviations during wavelet coefficient decomposition. Furthermore, this embodiment does not directly discard wavelet coefficients smaller than the threshold, thus preventing useful signals mixed in with noise from being filtered out, thereby further reducing deviations.

[0032] Meanwhile, this embodiment calculates different wavelet thresholds based on different decomposition levels, meaning the wavelet threshold is variable, thus achieving adaptive correction as the number of decomposition levels increases, thereby improving the adaptability of the wavelet threshold.

[0033] After constructing the initial wavelet threshold function, a signal denoising evaluation function can be built so that the optimal adjustment parameters of the initial wavelet threshold function can be found based on the denoising evaluation function.

[0034] In practical applications, the core assumption of this embodiment for constructing the signal denoising evaluation function is: an ideal wavelet threshold denoising process should decompose the original signal into a denoised signal (retaining all useful information) + residual (containing only unpredictable random noise). If this assumption holds, then the residual should ideally have two key characteristics: time-domain incorrelation (i.e., the ideal residual should exhibit a random distribution without obvious patterns or trends) and frequency-domain flatness (the ideal denoising process should retain as much useful information as possible from the original signal while separating out the noise (usually manifested as random, irregular signals). If the spectrum of the residual is flat, it indicates that the residual no longer contains the structured information of the original signal (such as periodicity, formants, etc.), suggesting that the denoising process may have effectively extracted the main features of the signal, and the remaining part is mainly random noise). Therefore, this embodiment constructs the signal denoising evaluation function based on the signal residual and from both the time and frequency domains. The process can be, but is not limited to, the steps S21 to S25 below.

[0035] S21. Establish the residual sequence between the power signal and the denoised power signal; in this embodiment, the residual sequence can be obtained by subtracting the denoised power signal from the power signal; at the same time, as mentioned above, the ideal residual has time-domain incorrelation, therefore, it can be subjected to multiple lag processing, so as to construct the lag correlation function based on the lag residual sequence, the process of which is shown in step S22 below.

[0036] S22. Perform multiple lag processing on the residual sequence to obtain several lagged residual sequences. In this embodiment, the number of lags can be, but is not limited to, the length of the residual sequence / 4 (rounded down if not an integer). For example, if the length of the residual sequence is 20, then the number of lags is 5, that is, a total of 5 lagged residual sequences are obtained from lag one, lag two, lag three, lag four, and lag five. After obtaining several lagged residual sequences, the autocorrelation coefficient can be calculated, as shown in step S23 below.

[0037] S23. Calculate the autocorrelation coefficient of each lagged residual sequence, and establish the lagged correlation function of the residual sequence based on each autocorrelation coefficient; in this embodiment, for any lagged residual sequence, first calculate its sample autocovariance, and then divide the sample autocovariance by the autocovariance of the original residual sequence (the autocovariance of the original residual sequence is essentially the variance) to obtain the autocorrelation coefficient of any lagged residual sequence.

[0038] Wherein, assuming the original residual sequence is W1, W2, ..., WN, then the formula for calculating the sample autocovariance of the v-th lagged residual sequence (i.e., lagged v period) is: ; In the formula, Let N represent the sample autocovariance of the v-th lag residual sequence, and N be the length of the residual sequence. This represents the d-th signal point in the original residual sequence. This represents the mean of the original residual sequence. This represents the (d+v)th signal point in the original residual sequence (which is essentially the signal point of the original residual sequence after a lag of v periods).

[0039] Thus, based on the aforementioned method, after calculating the autocorrelation coefficients of each lagged residual sequence, the lagged correlation function of the residual sequence can be established; wherein, the lagged correlation function is: ; In the formula, Indicates lagged correlation. Let V represent the autocorrelation coefficient of the v-th lagged residual sequence, and the entire formula represents the lagged correlation function, where V represents the number of lags.

[0040] Thus, after constructing the hysteresis correlation function, the spectral flatness function can be constructed from the frequency domain level, as shown in step S24 below.

[0041] S24. Establish the spectral flatness function of the residual sequence; in specific implementation, for example, but not limited to, performing an FFT transform on the residual sequence to obtain the power spectrum; and then, constructing the spectral flatness function based on the power spectrum; wherein, the spectral flatness function is the ratio of the geometric mean to the arithmetic mean of the power spectrum.

[0042] Thus, after constructing the spectral flatness function, the signal denoising evaluation function can be constructed by combining it with the hysteresis correlation function, as shown in step S25 below.

[0043] S25. Construct the signal denoising evaluation function using the hysteresis correlation function and the spectral flatness function; in this embodiment, for example, but not limited to, the following formula can be used to establish the aforementioned signal denoising evaluation function.

[0044] ; In the formula, This represents the signal denoising evaluation function. These represent the spectral flatness function and the hysteresis correlation function, respectively. These represent the spectral flatness weight and the hysteresis correlation weight, respectively.

[0045] In this embodiment, the closer the spectral flatness is to 1, the flatter the frequency is, and the better the denoising effect is. The closer it is to 0, the more peaks there are in the spectrum, and there may still be signals in the residuals, resulting in a poor denoising effect. Similarly, the smaller the hysteresis correlation, the weaker the autocorrelation of the sequence is, and the better the denoising effect is. Thus, the optimization objective of this evaluation function is to maximize the F value.

[0046] Thus, after constructing the signal denoising evaluation function based on the aforementioned steps S21 to S25, the optimization of the adjustment parameters can be performed based on this function, as shown in step S3 below.

[0047] S3. Based on the signal denoising evaluation function, the adjustment parameters in the initial wavelet threshold function are optimized to obtain the optimal adjustment parameters. Then, the initial wavelet threshold function is optimized using the optimal adjustment parameters to obtain the optimal wavelet threshold function. In specific applications, this embodiment adopts a swarm intelligence algorithm and uses the signal denoising evaluation function as the fitness function to optimize the adjustment parameters. The process is shown in steps S31 to S38 below.

[0048] S31. Obtain the individual population at the t-th iteration, where when t is 1, the individual population at the t-th iteration is the initial population. Each initial individual in the initial population is configured with a position vector and a velocity vector, and the position vector corresponding to any initial individual is used to represent a set of initial adjustment parameters. In this embodiment, the search space of three adjustment parameters can be determined first, including two morphological adjustment parameters (i.e., The search space for is [1, 10], while The search space is (0,1). Therefore, by randomly selecting values ​​within the aforementioned search space, multiple sets of initial adjustment parameters can be obtained. Then, these multiple sets of initial adjustment parameters can be used as position vectors of initial individuals. After initializing the velocity vector, an initial population can be constructed. Thus, after obtaining the individual population at the t-th iteration, the corresponding wavelet threshold function can be constructed based on each individual at the t-th iteration, and denoising can be performed. The process is shown in step S32 below.

[0049] S32. Based on each individual in the population at the t-th iteration, a wavelet threshold function corresponding to each individual is constructed, and the power signal is denoised using each wavelet threshold function to obtain the denoised power signal corresponding to each individual. In this embodiment, as previously explained, each individual corresponds to a set of adjustment parameters. Therefore, by substituting the adjustment parameters corresponding to each individual into the initial wavelet threshold function, the wavelet threshold function corresponding to each individual can be obtained. Then, the wavelet threshold function corresponding to each individual can be used to denoise the power signal to obtain the denoised power signal corresponding to each individual.

[0050] After obtaining the denoised power signals for each individual, the aforementioned signal denoising evaluation function can be used to calculate the individual fitness, as shown in step S33 below.

[0051] S33. Based on the signal denoising evaluation function, and using the power supply signal and the denoised power supply signal corresponding to each individual, calculate the fitness of each individual; in specific implementation, for any individual, calculate the residual sequence of the power supply signal and the denoised power supply signal corresponding to that individual, and then, using the aforementioned steps S22 to S25, calculate the signal denoising evaluation value F, which is its fitness (i.e., the larger the fitness of any individual, the better the denoising performance); thus, after calculating the fitness of each individual based on the aforementioned method, the selection of the globally optimal individual can be performed, as shown in step S34 below.

[0052] S34. Based on the fitness of each individual, determine the globally optimal individual at the t-th iteration. In this embodiment, first determine the maximum fitness from the fitness of each individual at the t-th iteration; then, determine whether the maximum fitness is greater than the globally optimal individual at the (t-1)-th iteration. If so, take the individual corresponding to the maximum fitness as the globally optimal individual at the t-th iteration; otherwise, take the globally optimal individual at the (t-1)-th iteration as the globally optimal individual at the t-th iteration. Of course, when t is 1, the globally optimal individual at the first iteration is the individual with the maximum fitness in the population at the first iteration.

[0053] After determining the globally optimal individual at the t-th iteration, it can be determined whether the iteration stopping condition is met, as shown in step S35 below.

[0054] S35. Determine whether the iteration stopping condition is met; in specific implementation, the iteration stopping condition may be, but is not limited to, t reaching the maximum number of iterations, or the fitness of the globally optimal individual being greater than or equal to the fitness threshold; if the aforementioned iteration stopping condition is not met, then population update is required, the process of which is shown in step S36 below.

[0055] S36. If not, a dual-modal update mechanism is adopted to update the velocity vector of each individual at the t-th iteration, thereby obtaining the updated velocity vector corresponding to each individual. In specific implementation, in the traditional particle swarm optimization algorithm, individuals continuously move towards the direction of their historical optimum and the global optimum to find the optimal solution. However, once an individual finds a local optimum during its movement, it is difficult to escape and may attract other individuals to gather at that point, causing the population to fall into a local optimum. At the same time, as other individuals get closer and closer to the global optimum, the diversity of the population will decrease, resulting in a significant decrease in the convergence speed, or even evolutionary stagnation. Therefore, to solve the aforementioned shortcomings, this embodiment proposes a dual-modal update mechanism, the process of which is shown in the following steps S36a to S36d.

[0056] S36a. Calculate the search weight and search control factor at the t-th iteration; in specific implementation, the following formula can be used, but is not limited to, to calculate the aforementioned search weight.

[0057] ; In the formula, Let be the search weight at the t-th iteration. These represent the maximum search weight and the minimum search weight, respectively. All are weighting control coefficients.

[0058] As can be seen from the aforementioned formula, this embodiment uses a composite nonlinear function (nested exponential and logarithmic functions) to control the decay process of the search weight. Compared with the traditional linear decrease, it can provide a more complex search curve. In particular, the exponential function term in the formula has a large initial value in the iteration, and through nested fractional and logarithmic operations, the search weight can be maintained near the maximum search weight for a long time in the early stage of the iteration. This gives the individual sufficient time to perform a global search and search different regions extensively, which helps to discover potential optimal regions. In the later stage of the iteration, the search weight is gradually reduced. Due to the sufficient exploration in the early and middle stages, the algorithm has already located a good region. At this time, the smaller search weight is conducive to the individual to carry out fine local development in this region and accelerate convergence to the optimal solution.

[0059] After calculating the search weight at the t-th iteration based on the aforementioned formula, the search control factor at the t-th iteration can be calculated, and the calculation formula is as follows: ; In the formula, Both represent the search control factors at the t-th iteration. This indicates the maximum number of iterations.

[0060] Thus, after calculating the search control factor at the t-th iteration based on the aforementioned formula, the position of the individual can be updated, as shown in steps S36b to S36d below.

[0061] S36b. For any individual in the population at the t-th iteration, generate the search probability of that individual; in this embodiment, a random number can be randomly generated between [0,1] as the search probability of that individual; after obtaining the search probability of that individual, the speed update method of that individual can be determined based on its magnitude with the probability threshold, as shown in steps S36c and S36d below.

[0062] S36c. Determine whether the search probability is less than the probability threshold. In specific implementation, the probability threshold can be, but is not limited to, set to 0.3. When the search probability is less than 0.3, a perturbation global update mechanism is used to update the speed. Conversely, a local update mechanism is used to update the speed. The process is shown in step S36d below.

[0063] S36d. If yes, then based on the search weight and search control factor, and using a perturbation global update mechanism, update the velocity vector of any individual to obtain the updated velocity vector corresponding to that individual; otherwise, based on the search weight and search control factor, and using a local update mechanism, update the velocity vector of any individual to obtain the updated velocity vector corresponding to that individual, so that after all individuals at the t-th iteration have been polled, the updated velocity vectors corresponding to each individual are obtained.

[0064] In practical implementation, when a perturbation-based global update mechanism is adopted, the update formula for the velocity vector of any individual is as follows: ; In the formula, This represents the velocity vector of any of the individuals. Indicates the updated , This represents the search weight at the t-th iteration. Both represent the search control factors at the t-th iteration. This represents the historical best position vector of any given individual. This represents the position vector of any of the individuals. Let represent the position vector corresponding to the globally optimal individual at the t-th iteration. Scaling factor , It is a random number between [0,1].

[0065] Similarly, when using a local update mechanism, the update formula for the velocity vector of any individual is: ; In the formula, The oscillation coefficient is, and .

[0066] As can be seen from the above speed update formula, this embodiment divides the speed update into two modes based on the search probability. When the search probability is less than 0.3, the individual flies towards the historical optimal and the globally optimal individual. However, the reference point of the individual is not its current position, but the position after introducing a scaling factor. Thus, this can be regarded as a perturbation-based global exploration, which helps the individual to search extensively in the early stage. When the search probability is less than 0.3, its speed update formula introduces cos2πη, which allows the individual to perform a spiral search around the optimal position, thereby allowing the individual to perform a local fine scan near the optimal solution. Based on this, this dual-modal mechanism prevents the individual from flying solely towards the optimal solution. It balances the contradiction between global exploration and local development in the algorithm. That is, global search ensures that the individual does not gather too early, while spiral motion ensures that a high-precision search can be performed after discovering a potential optimal region.

[0067] Meanwhile, this embodiment introduces in both cases... This term is essentially a direction control factor, which alternates between 1 and -1 with the number of iterations. This means that while the individual is learning towards the optimal position, its direction of movement will periodically reverse, thus achieving an oscillating search. This back-and-forth oscillation mechanism greatly enriches the diversity of the individual's flight direction. Thus, when the individual has approached a local optimum, the direction reversal characteristic may also throw it out of that area, thereby significantly enhancing the individual's ability to escape local optima.

[0068] Simultaneously, the introduction of a cosine factor and random scaling enriches the diversity of the population. Specifically, the introduction of the cosine function transforms the particle's trajectory from a simple straight line into a complex curve with periodic oscillations, significantly increasing the possibility of particles exploring unknown spaces. Furthermore, combined with the aforementioned formula for calculating the search control factor, the search control factor provided in this embodiment changes with the number of iterations. Specifically, in the early stages of iteration, sin(tπ / 2T) is close to 0. Larger Smaller individuals tend to learn from their own historical experiences (stronger self-awareness), which helps maintain individual diversity, allowing them to explore in a broader space and avoiding premature clustering towards a local optimum. However, in the later stages of iteration... Smaller The smaller the value, the more inclined individuals are to learn towards the global optimum. This is beneficial for individuals to conduct fine-grained searches around the discovered global optimum region, accelerating convergence. Furthermore, the introduction of the sine function makes the changes of the two search control factors smooth and non-linear, thus enabling the algorithm to balance convergence speed and final accuracy.

[0069] After the velocity update of each individual is completed through the aforementioned bimodal update mechanism, the position update can be performed, as shown in step S37 below.

[0070] S37. Using the updated velocity vectors corresponding to each individual, update the position vectors of each individual to obtain the individual population at the (t+1)th iteration; in this embodiment, the traditional particle swarm position update method is used to update the position vectors of each individual, and the principle will not be elaborated here.

[0071] After updating the position vectors of each individual, the population of individuals for the next iteration can be obtained. At this point, the aforementioned iterative process can be repeated until the iteration stopping condition is met, at which point the optimal adjustment parameters can be obtained. The iterative process is as shown in step S38 below.

[0072] S38. Increment t by 1 and reacquire the individual population at the t-th iteration until the iteration stopping condition is met, so as to determine the optimal adjustment parameter based on the globally optimal individual at the iteration stopping condition.

[0073] Therefore, after finding the optimal adjustment parameters through the aforementioned steps S31 to S38, the optimal adjustment parameters can be substituted into the aforementioned initial wavelet threshold function to obtain the optimal wavelet threshold function; then, based on this, the power signal can be denoised, as shown in step S4 below.

[0074] S4. The power signal is denoised using the optimal wavelet threshold function to obtain an initial denoised signal. In this embodiment, wavelet denoising is a commonly used denoising method, and its principle will not be elaborated here.

[0075] Thus, after obtaining the initial denoised signal, secondary denoising can be performed, namely, the removal of impulse noise, as shown in step S5 below.

[0076] S5. A hybrid morphological filtering algorithm is used to perform morphological filtering on the initial denoised signal to obtain a denoised signal. In specific implementation, for example, but not limited to, the following steps S51 to S54 can be used to perform morphological filtering on the initial denoised signal.

[0077] S51. Obtain the structure element sequence; in specific implementation, the structure element is usually a flat straight line or triangle, and the length can be determined according to the pulse width; in this way, after obtaining the structure element sequence, mixed morphological filtering can be performed, and the process is shown in steps S52 to S54 below.

[0078] S52. Based on the sequence of structuring elements, the initial denoised signal is sequentially subjected to opening and closing operations to obtain the first filtered signal.

[0079] S53. Based on the sequence of structuring elements, the initial denoised signal is sequentially subjected to closing and opening operations to obtain the second filtered signal.

[0080] In practical implementation, a single opening or closing operation can only suppress pulses in one direction. However, in actual power signals, positive and negative pulses coexist. Therefore, a mixed and alternating processing strategy of opening and closing operations is adopted to suppress bidirectional pulses simultaneously. In this way, after obtaining two filtered signals, they can be averaged to balance the filtering effects of the two orders, avoid phase shift, and make the signal smoother. The process is shown in step S54 below.

[0081] S54. The first filtered signal and the second filtered signal are averaged to obtain a denoised signal after averaging. In this embodiment, the first filtered signal is f1 and the second filtered signal is f2. Then, the denoised signal is (f1+f2) / 2.

[0082] After the pulse noise is filtered out through the aforementioned steps S51 to S54, fault features can be extracted so that power supply anomaly identification can be performed based on the extracted fault features; wherein, the fault feature extraction process is as shown in step S6 below.

[0083] S6. Perform fault feature extraction processing on the denoised signal to obtain power fault features; in specific implementation, for example, but not limited to, the following steps S61 to S69 can be used to extract power fault features.

[0084] S61. Obtain the optimal wavelet decomposition level and use the optimal wavelet decomposition level to perform wavelet decomposition processing on the denoised signal to obtain wavelet signals at different decomposition scales. In specific implementation, examples include, but are not limited to, using experimental verification methods based on evaluation indicators (such as based on indicators such as signal-to-noise ratio and root mean square error) to predetermine the optimal wavelet decomposition level, and then using the optimal wavelet decomposition level to perform wavelet decomposition on the denoised signal. In this embodiment, the optimal wavelet decomposition level is preferably 3. Of course, different decomposition levels can be selected according to actual use, and this embodiment is not limited to the above examples.

[0085] After wavelet decomposition of the denoised signal is completed, frequency domain features can be extracted, as shown in steps S62 to S67 below.

[0086] S62. Extract the wavelet signal at the largest decomposition scale from the wavelet signals at all decomposition scales, and obtain all frequency band sequences in the extracted wavelet signal. In this embodiment, a signal with a sampling frequency of fs, after i-level wavelet decomposition, has 2^i frequency bands at the i-th decomposition scale, and the width of each frequency band is fs / (2^(i+1)). Therefore, there are two frequency bands at the first decomposition scale with a bandwidth of fs / 4, four frequency bands at the second decomposition scale with a bandwidth of fs / 8, and eight frequency bands at the third decomposition scale with a bandwidth of fs / 16. Among them, the frequency bands at the first two decomposition scales are too wide, resulting in insufficient resolution, that is, harmonics may be in the same frequency band, leading to feature ambiguity. Therefore, in this embodiment, all frequency band sequences in the wavelet signal at the third decomposition scale are selected for frequency domain feature extraction, and the process is shown in steps S63 to S67 below.

[0087] S63. For any frequency band sequence, the frequency band sequence is divided into several sub-frequency bands. In this embodiment, for example, but not limited to, any frequency band sequence can be divided into two equal parts to obtain two sub-frequency bands. The reason for dividing the frequency band is that, during operation, the power module has various nonlinear loads, such as electronic devices and strong magnetic loads (computers, UPS, etc.), which generate harmonic current and voltage signals with frequencies higher than the fundamental frequency. These signals are superimposed on the fundamental frequency signal to form power grid harmonics, causing the voltage and current signals to become non-stationary time-varying signals. Therefore, the power signal may be a non-stationary time-varying signal, and the traditional method of using the total energy of the entire frequency band cannot describe the time-varying nature of the signal. Therefore, this embodiment divides the frequency band into equal parts, and by calculating the energy of the sub-frequency bands, the local energy is used as a characteristic parameter for anomaly identification.

[0088] Thus, after the equal division of any frequency sequence is completed, the distribution dispersion of each sub-band can be determined, as shown in step S64 below.

[0089] S64. Based on each sub-band, calculate the distribution dispersion of each sub-band; in this embodiment, different local energies have different characterization effects on power supply anomalies, that is, different features have different sensitivities to power supply anomalies. Therefore, it is necessary to extract the local energy of the sub-bands that are sensitive to power supply anomalies as key features; based on this, this embodiment introduces the distribution dispersion to measure the sensitivity of the sub-bands to power supply anomalies.

[0090] Optionally, the calculation process for the distribution dispersion of any sub-band can be as follows: calculate the mean and standard deviation of each signal point in the sub-band; then, divide the standard deviation by the mean to obtain the distribution dispersion of the sub-band.

[0091] Thus, after calculating the distribution dispersion of each sub-band, the sub-bands that are sensitive to power supply anomalies can be extracted based on this, as shown in step S65 below.

[0092] S65. Based on the distribution dispersion of each sub-band, extract the sub-bands used for power supply fault diagnosis from several sub-bands as key sub-bands; in specific applications, this embodiment uses sub-bands with a distribution dispersion greater than 0.15 as key sub-bands; thus, after extracting the key sub-bands, the frequency band energy can be calculated, as shown in step S66 below.

[0093] S66. Calculate the energy of each key sub-band, and after polling all frequency band sequences, obtain the energy of all key sub-bands in each frequency band sequence; in this embodiment, the energy of a key sub-band is the sum of the squares of the decomposition coefficients (i.e., wavelet decomposition coefficients) within that key sub-band. Thus, process the remaining frequency band sequences in the aforementioned manner, and after polling all frequency band sequences, obtain the energy of several key sub-bands. Finally, based on this, the frequency domain characteristics of the power signal can be formed, as shown in step S67 below.

[0094] S67. The frequency domain characteristics of the power signal are formed by using the energy of all key sub-bands in each frequency band sequence; in this embodiment, the frequency domain characteristics can be obtained by splicing the energy of all key sub-bands into a vector.

[0095] After obtaining the frequency domain features, the time domain features can be extracted, as shown in step S68 below.

[0096] S68. Perform time-domain feature extraction processing on the power supply signal to obtain the time-domain features of the power supply module; extract the rate of change of the power supply signal at adjacent time points (i.e., the rate of change of voltage and the rate of change of current) to form a rate of change vector; at the same time, calculate the average value of the power supply signal, then calculate the absolute value of the difference between each signal point in the power supply signal and the average value, and divide the absolute value of each difference by the average value to obtain the deviation of the power supply signal (i.e., the voltage and current deviation at each time point), and use the deviation of the power supply signal to form a deviation vector; finally, the aforementioned time-domain features can be formed using the rate of change vector and the deviation vector.

[0097] Thus, after extracting the frequency and time domain characteristics of the power signal, power fault characteristics can be formed, as shown in step S69 below.

[0098] S69. The power supply fault characteristics are formed by using the frequency domain characteristics and the time domain characteristics.

[0099] After extracting the power fault characteristics of the power module through the aforementioned steps S61 to S69, power abnormality diagnosis can be performed based on these characteristics, as shown in step S7 below.

[0100] S7. Input the power fault features into the power fault detection model to obtain the abnormal diagnosis result of the power module. In this embodiment, the power fault detection model can be, but is not limited to, a trained SVM model, that is, it is trained by taking the sample power fault features of several sample power modules as input and the fault identification result of each sample power module as output (i.e., the abnormal diagnosis result). The abnormality type of the power module can be either abnormal or the power module is operating normally, i.e., binary classification is performed. Of course, multi-class classification can also be performed, that is, outputting the specific fault type, such as short circuit, overvoltage, undervoltage, open circuit, etc.

[0101] Therefore, through the power module anomaly diagnosis method described in detail in steps S1 to S7 above, this invention constructs a signal denoising evaluation function and optimizes the adjustment parameters in the initial wavelet threshold function to obtain objective and global optimal adjustment parameters. This improvement completely eliminates the subjectivity and uncertainty of traditional techniques that rely on human experience to set initial values, and solves the problem of denoising results varying from person to person due to differences in the experience of different operators. It ensures the objectivity and stability of the signal preprocessing stage and lays a solid foundation for the accuracy of subsequent fault diagnosis. At the same time, based on the optimal wavelet threshold function denoising, this invention further introduces a hybrid morphological filtering algorithm to process the initial denoised signal. This combined strategy effectively overcomes the limitation of traditional wavelet denoising, which is only good at processing white noise, and can effectively filter out impulse noise in the electromagnetic environment, achieving deep suppression of complex mixed noise. Thus, through objective parameter optimization and hybrid filtering strategies, this invention significantly improves the denoising quality and robustness of power signals, thereby making the extracted power fault features more accurate and ultimately greatly improving the accuracy of power module anomaly diagnosis.

[0102] like Figure 2 As shown, the second aspect of this embodiment provides a hardware system for implementing the power module anomaly diagnosis method described in the first aspect of the embodiment, comprising: The acquisition unit is used to acquire the power signal of the power module.

[0103] The construction unit is used to construct the initial wavelet threshold function of the power signal and the signal denoising evaluation function based on the time-frequency domain.

[0104] The parameter optimization unit is used to optimize the adjustment parameters in the initial wavelet threshold function based on the signal denoising evaluation function to obtain the optimal adjustment parameters, and then use the optimal adjustment parameters to optimize the initial wavelet threshold function to obtain the optimal wavelet threshold function.

[0105] The denoising unit is used to denoise the power signal using the optimal wavelet threshold function to obtain the initial denoised signal. The denoising unit is also used to perform morphological filtering on the initial denoising signal using a hybrid morphological filtering algorithm to obtain the denoised signal.

[0106] The feature extraction unit is used to perform fault feature extraction processing on the denoised signal to obtain power fault features.

[0107] An anomaly diagnosis unit is used to input the power failure characteristics into the power failure detection model to obtain the anomaly diagnosis results of the power module.

[0108] 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.

[0109] like Figure 3 As shown, the third aspect of this embodiment provides a power module anomaly diagnosis device. 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 power module anomaly diagnosis method as described in the first aspect of the embodiment.

[0110] 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.

[0111] 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.

[0112] 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.

[0113] The fourth aspect of this embodiment provides a storage medium that stores instructions containing the power module anomaly diagnosis method described in the first aspect of the embodiment. That is, the storage medium stores instructions that, when executed on a computer, perform the power module anomaly diagnosis method as described in the first aspect of the embodiment.

[0114] 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.

[0115] 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.

[0116] 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 power module anomaly diagnosis method 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.

[0117] 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 for diagnosing power module anomalies, characterized in that, include: Obtain the power signal from the power module; An initial wavelet threshold function for the power supply signal and a signal denoising evaluation function based on the time-frequency domain are constructed. Based on the signal denoising evaluation function, the adjustment parameters in the initial wavelet threshold function are optimized to obtain the optimal adjustment parameters. Then, the initial wavelet threshold function is optimized using the optimal adjustment parameters to obtain the optimal wavelet threshold function. The power signal is denoised using the optimal wavelet threshold function to obtain an initial denoised signal. A hybrid morphological filtering algorithm is used to perform morphological filtering on the initial denoised signal to obtain the denoised signal. The denoised signal is subjected to fault feature extraction processing to obtain power supply fault features; The power failure characteristics are input into the power failure detection model to obtain the abnormal diagnosis results of the power module.

2. The method according to claim 1, characterized in that, Construct a signal denoising evaluation function, including: Establish the residual sequence between the power signal and the denoised power signal; The residual sequence is subjected to multiple lag processes to obtain several lag residual sequences; Calculate the autocorrelation coefficient of each lagged residual sequence, and establish the lagged correlation function of the residual sequence based on each autocorrelation coefficient; Establish the spectral flatness function of the residual sequence; The signal denoising evaluation function is constructed using the hysteresis correlation function and the spectral flatness function.

3. The method according to claim 1, characterized in that, Based on the signal denoising evaluation function, the adjustment parameters in the initial wavelet threshold function are optimized to obtain the optimal adjustment parameters, including: Obtain the individual population at the t-th iteration, where when t is 1, the individual population at the t-th iteration is the initial population. Each initial individual in the initial population is configured with a position vector and a velocity vector, and the position vector corresponding to any initial individual is used to represent a set of initial adjustment parameters. Based on each individual in the population at the t-th iteration, a wavelet threshold function corresponding to each individual is constructed, and the power signal is denoised using each wavelet threshold function to obtain the denoised power signal corresponding to each individual. The fitness of each individual is calculated based on the signal denoising evaluation function and using the power supply signal and the denoised power supply signal corresponding to each individual. Based on the fitness of each individual, the globally optimal individual at the t-th iteration is determined; Determine if the iteration stopping condition is met; If not, a dual-modal update mechanism is used to update the velocity vector of each individual at the t-th iteration, thus obtaining the updated velocity vector of each individual. The updated velocity vectors of each individual are used to update the position vectors of each individual to obtain the population of individuals at the (t+1)th iteration. Increment t by 1 and reacquire the individual population at the t-th iteration until the iteration stopping condition is met. Based on the globally optimal individual at the iteration stopping condition, determine the optimal adjustment parameter.

4. The method according to claim 3, characterized in that, A dual-modal update mechanism is used to update the velocity vectors of each individual at the t-th iteration, including: Calculate the search weight and search control factor at the t-th iteration; For any individual in the population at the t-th iteration, the search probability of generating that individual; Determine if the search probability is less than the probability threshold; If so, the velocity vector of any individual is updated based on the search weight and search control factor, using a perturbation global update mechanism, to obtain the updated velocity vector corresponding to that individual. Otherwise, the velocity vector of any individual is updated based on the search weight and search control factor, using a local update mechanism, to obtain the updated velocity vector corresponding to that individual. This process is repeated until all individuals at the t-th iteration have been polled, at which point the updated velocity vectors for each individual are obtained.

5. The method according to claim 4, characterized in that, Based on search weights and search control factors, and employing a perturbation-based global update mechanism, the velocity vector of any individual is updated, including: Update the velocity vector of any individual according to the following formula; ； In the formula, This represents the velocity vector of any of the individuals. Indicates the updated , This represents the search weight at the t-th iteration. Both represent the search control factors at the t-th iteration. This represents the historical best position vector of any given individual. This represents the position vector of any of the individuals. Let represent the position vector corresponding to the globally optimal individual at the t-th iteration. Scaling factor , A random number between [0, 1]; Accordingly, based on search weights and search control factors, and employing a local update mechanism, the velocity vector of any individual is updated, including: Update the velocity vector of any individual according to the following formula; ； In the formula, The oscillation coefficient is, and .

6. The method according to claim 1, characterized in that, A hybrid morphological filtering algorithm is used to perform morphological filtering on the initial denoised signal to obtain a denoised signal, including: Obtain the sequence of structure elements; Based on the structuring element sequence, the initial denoised signal is sequentially subjected to opening and closing operations to obtain the first filtered signal; Based on the structuring element sequence, the initial denoised signal is sequentially subjected to closing and opening operations to obtain the second filtered signal; The first filtered signal and the second filtered signal are averaged to obtain the denoised signal.

7. The method according to claim 1, characterized in that, The denoised signal is processed to extract fault features, resulting in power supply fault features, including: The optimal wavelet decomposition level is obtained, and the denoised signal is processed by wavelet decomposition using the optimal wavelet decomposition level to obtain wavelet signals at different decomposition scales. From the wavelet signals at all decomposition scales, extract the wavelet signal at the largest decomposition scale, and obtain all frequency band sequences in the extracted wavelet signal; For any frequency band sequence, the sequence is divided into several sub-frequency bands. Based on each sub-band, the distribution dispersion of each sub-band is calculated; Based on the distribution dispersion of each sub-band, the sub-band used for power supply fault diagnosis is extracted from several sub-bands and used as the key sub-band. The energy of each key sub-band is calculated, and after polling all frequency band sequences, the energy of all key sub-bands in each frequency band sequence is obtained. The frequency domain characteristics of the power signal are formed by utilizing the energy of all key sub-bands in each frequency band sequence. The power signal is processed by time-domain feature extraction to obtain the time-domain features of the power module; The power supply fault characteristics are formed by using the frequency domain characteristics and the time domain characteristics.

8. A power module fault diagnosis system, characterized in that, include: Acquisition unit, used to acquire power signals from the power module; The construction unit is used to construct the initial wavelet threshold function of the power signal and the signal denoising evaluation function based on the time and frequency domain; The parameter optimization unit is used to optimize the adjustment parameters in the initial wavelet threshold function based on the signal denoising evaluation function to obtain the optimal adjustment parameters, and to optimize the initial wavelet threshold function using the optimal adjustment parameters to obtain the optimal wavelet threshold function. The denoising unit is used to denoise the power signal using the optimal wavelet threshold function to obtain the initial denoised signal. The denoising unit is also used to perform morphological filtering on the initial denoising signal using a hybrid morphological filtering algorithm to obtain the denoised signal. The feature extraction unit is used to perform fault feature extraction processing on the denoised signal to obtain power fault features; An anomaly diagnosis unit is used to input the power failure characteristics into the power failure detection model to obtain the anomaly diagnosis results of the power module.

9. An electronic device, characterized in that, include: A memory, a processor, and a transceiver are sequentially connected in communication, wherein the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the power module abnormality diagnosis method as described in any one of claims 1 to 7.

10. A computer program product containing instructions, characterized in that, When the instruction is executed on the computer, it causes the computer to perform the power module abnormality diagnosis method as described in any one of claims 1 to 7.

Images