Self-adaptive stochastic resonance weak fault detection method based on kurtosis optimization

Abstract

The invention provides an adaptive stochastic resonance weak fault detection method based on kurtosis optimization, and belongs to the technical field of fault detection, and the method comprises the steps: collecting and preprocessing a transient signal on a cable line; constructing a bistable stochastic resonance model; kurtosis of the output signal is used as a fitness function, and potential well parameters of the bistable stochastic resonance model are optimized based on a particle swarm optimization algorithm; and substituting the optimized potential well parameter into the bistable stochastic resonance model, and processing the preprocessed transient signal to obtain an enhanced output signal for fault detection. The method has the beneficial effects that the stochastic resonance system is constructed on the basis of the classical bistable model, the kurtosis of the output signal is used as the fitness function of the particle swarm optimization algorithm, the potential well parameters are adaptively optimized, the potential well parameters are driven to automatically converge to the optimal state, and the robustness of the system is improved. The high-impedance grounding fault traveling wave signal can be extracted and enhanced from a strong noise background, and the detection sensitivity and reliability are improved.

Description

Technical Field

[0001] This invention relates to the field of fault detection technology, and in particular to an adaptive stochastic resonance weak fault detection method based on kurtosis optimization. Background Technology

[0002] High-impedance fault (HIF) is a common but extremely difficult-to-detect type of fault in the operation and maintenance of high-voltage cable lines. This type of fault is usually caused by minor damage or aging of the cable insulation, or by the instantaneous contact of external objects (such as tree branches or contaminants). The transition resistance at the grounding point can be as high as several thousand ohms, resulting in a very small fault current and weak fault energy.

[0003] When a high-impedance ground fault occurs, a traveling wave signal is generated that propagates bidirectionally along the cable line. However, due to the weak fault energy, the initial amplitude of the traveling wave signal is extremely low, and it is easily affected by the characteristics of the line itself and various electromagnetic environmental noises during propagation. As a result, the signal is completely submerged in strong background noise when it reaches the detection end, and the signal-to-noise ratio (SNR) is extremely low.

[0004] Currently, existing fault detection methods, such as those based on current surges, wavelet transforms, or Fourier analysis, have significant limitations when dealing with weak signals with extremely low signal-to-noise ratios. Traditional threshold detection methods struggle to set a suitable threshold in strong noise environments, requiring both reliable detection of weak fault signals and avoidance of false alarms due to noise interference, resulting in insufficient detection sensitivity. Because they cannot effectively separate the fault signal from the noise, conventional methods often misinterpret weak fault traveling wave signals as normal ambient noise, leading to missed fault detection and increasing the false alarm rate. Furthermore, missed fault detection prevents the triggering of fault location devices, hindering timely fault clearance. Persistent high-impedance grounding faults can further develop into permanent and more destructive short-circuit faults, causing serious damage to electrical equipment and even posing a significant threat to the personal safety of surrounding personnel, presenting a substantial safety hazard. Summary of the Invention

[0005] To address the above technical problems, this invention provides an adaptive stochastic resonance weak fault detection method based on kurtosis optimization. This method aims to solve the problems of weak traveling wave signals, extremely low signal-to-noise ratio, and difficulty in detection caused by high impedance grounding faults in high-voltage cables.

[0006] The technical problem solved by this invention can be achieved by the following technical solution: an adaptive stochastic resonance weak fault detection method based on kurtosis optimization, comprising: step S1, acquiring transient signals on a cable line and preprocessing the transient signals to obtain preprocessed transient signals; step S2, constructing a bistable stochastic resonance model based on the Langevin equation, and numerically solving the Langevin equation using the fourth-order Runge-Kutta algorithm to obtain an output signal; step S3, using the kurtosis of the output signal as a fitness function, optimizing the potential well parameters of the bistable stochastic resonance model based on the particle swarm optimization algorithm to obtain optimized potential well parameters; step S4, substituting the optimized potential well parameters into the bistable stochastic resonance model, processing the preprocessed transient signals to obtain an enhanced output signal; and step S5, performing fault detection based on the enhanced output signal to determine whether a fault has occurred.

[0007] Preferably, in step S1, the preprocessing includes normalization preprocessing, which employs the following formula:

[0008] in, Indicates a transient signal; n Represents discrete time points; This represents the minimum value of the transient signal within the sliding data window; This represents the maximum value of the transient signal within the sliding data window; This represents the preprocessed transient signal.

[0009] Preferably, in step S2, the differential form of the Langevin equation is:

[0010] in, x This represents the output signal of a random resonance system; a , b ... This represents the preprocessed transient signal; This represents the differential form of the Langevin equation.

[0011] Preferably, in step S2, the output signal is:

[0012] in, , These represent the output signals, n、n +1 represents a discrete time point; a , b ... This represents the preprocessed transient signal; h This indicates the calculation step size.

[0013] Preferably, step S3 includes: step S31, initializing the particle swarm; step S32, for each particle, substituting the particle into the bistable stochastic resonance model to obtain an output signal obtained by numerically solving the preprocessed transient signal, and using the kurtosis of the output signal as the fitness value of the particle; step S33, updating the individual optimum and global optimum based on the fitness value of the particle; step S34, updating the particle velocity and position; step S35, repeating steps S32-S34 until the particle swarm converges or reaches a preset number of iterations, and outputting the optimized potential well parameters.

[0014] Preferably, step S33 includes: comparing the fitness value of the particle with the historical best fitness value of the corresponding particle; if the fitness value of the particle exceeds the historical best fitness value, then updating the historical best fitness value to the fitness value of the particle and recording the corresponding position; comparing the fitness value of the particle with the global best fitness value; if the fitness value of the particle exceeds the global best fitness value, then updating the global best fitness value to the fitness value of the particle and recording the corresponding position.

[0015] Preferably, in step S34, the particle velocity and position are updated using the following formula:

[0016]

[0017] in, Indicates inertia weight; This indicates the particle velocity before the update; This indicates the updated particle velocity; , These represent learning factors; , These represent random numbers between [0, 1]. Indicates the optimal position for an individual; Indicates the particle position before the update; Indicates the globally optimal position; Indicates the updated particle position, subscript i Indicates particle index, d Indicates the dimension of the potential well parameters. t、t +1 represents the number of iterations.

[0018] Preferably, step S5 includes: comparing the enhanced output signal with a preset threshold; when the absolute value of the enhanced output signal exceeds the preset threshold, determining that a fault has occurred, and recording the fault occurrence timestamp information.

[0019] Preferably, before comparing the enhanced output signal with the preset threshold, the method further includes setting the preset threshold based on the optimized potential well parameters.

[0020] Preferably, the preset threshold is:

[0021] in, Indicates the safety factor; , These represent the optimized potential well parameters; This indicates a preset threshold.

[0022] The advantages or beneficial effects of the technical solution of the present invention are as follows: The present invention constructs a stochastic resonance system based on the classical bistable model and uses a particle swarm optimization algorithm to adaptively optimize the key parameters of the system. In particular, the kurtosis of the system output signal is used as the fitness function of the particle swarm optimization algorithm to drive the system parameters to automatically converge to the optimal state. This enables the effective extraction and enhancement of the high impedance grounding fault traveling wave signal from the background of strong noise, thereby improving the sensitivity and reliability of detection. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the adaptive stochastic resonance weak fault detection method based on kurtosis optimization in a preferred embodiment of the present invention. Figure 2 This is a flowchart illustrating the particle swarm optimization algorithm in a preferred embodiment of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0026] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0027] In a preferred embodiment of the present invention, based on the above-mentioned problems existing in the prior art, an adaptive stochastic resonance weak fault detection method based on kurtosis optimization is provided. The core idea is to construct a stochastic resonance system based on the classical bistable model and use the particle swarm optimization (PSO) algorithm to adaptively optimize the key parameters of the system. In particular, the present invention uses the kurtosis value of the output signal of the stochastic resonance (SR) system as the fitness function of the particle swarm optimization (PSO) algorithm to drive the system parameters to automatically converge to the optimal state, thereby maximizing the enhancement of the transient traveling wave fault signal submerged in strong noise.

[0028] like Figure 1 As shown, the method includes the following steps: Step S1, acquiring transient signals on the cable line and preprocessing the transient signals to obtain preprocessed transient signals; specifically, the transient signals can be obtained through a signal acquisition module. The signal acquisition module is electrically connected to a high-frequency current transformer (HFCT) or a coupling capacitance sensor installed at both ends of the cable line or at designated monitoring points.

[0029] The signal acquisition module acquires transient signals on the cable line at a high sampling rate (e.g., ≥10MHz). These transient signals are discrete digital signal sequences, denoted as... .

[0030] Since stochastic resonance systems have certain requirements on the amplitude of the input signal, generally speaking, to meet the small parameter condition, the signal amplitude is usually required to be less than 1. Therefore, the acquired raw signal needs to be normalized preprocessed.

[0031] Specifically, in step S1, the preprocessing includes normalization preprocessing, which uses the following formula:

[0032] in, Indicates a transient signal; n Represents discrete time points; This represents the minimum value of the transient signal within the sliding data window; This represents the maximum value of the transient signal within the sliding data window; This represents the preprocessed transient signal.

[0033] This invention employs a linear normalization method to normalize the input transient signal. The signal is scaled to the interval [-1, 1] to obtain the preprocessed transient signal. .

[0034] Step S2: Construct a bistable stochastic resonance model based on the Langevin equation, and use the fourth-order Runge-Kutta algorithm to numerically solve the Langevin equation to obtain the output signal; Specifically, this embodiment of the invention uses the classical bistable Langevin equation as the mathematical model for a stochastic resonance (SR) system. This model describes the motion of a Brownian particle in a double potential well, and its dynamic equation is as follows:

[0035] in, Represent the differential form of the Langevin equation; x The output signal of a stochastic resonance system is time. t The function, denoted as , representing the displacement of the particle; This represents the preprocessed transient signal sequence, which includes weak fault traveling wave signals and background noise. a , b Let represent the potential well parameters of the bistable stochastic resonance model, respectively. These two parameters together determine the shape of the potential well, including its depth and width, which are key to the performance of the stochastic resonance system and are also the objects that this invention needs to adaptively optimize.

[0036] For discrete signals, the above differential equations can be solved numerically using the fourth-order Runge-Kutta method to obtain the output signal. :

[0037] in, , These represent the output signals, n、n +1 represents a discrete time point; a , b ... This represents the preprocessed transient signal; h The step size is typically the reciprocal of the sampling period. , The sampling frequency.

[0038] Step S3: Using the kurtosis of the output signal as the fitness function, the potential well parameters of the bistable stochastic resonance model are optimized using the Particle Swarm Optimization (PSO) algorithm to obtain the optimized potential well parameters. Specifically, to achieve the best signal enhancement effect in the stochastic resonance (SR) system, the optimal combination of parameters must be found. ,in They represent the potential well parameters respectively. a , b The optimized value.

[0039] This invention proposes a parameter adaptive optimization based on kurtosis optimization, using the particle swarm optimization (PSO) algorithm and employing the kurtosis of the output signal as the fitness function.

[0040] Kurtosis is a dimensionless parameter that measures the steepness or flatness of a signal's probability density distribution curve, and it is highly sensitive to impulsive components in the signal. Its calculation formula is:

[0041] in, This represents the output signal of a stochastic resonance (SR) system. Indicates the sequence length; This represents the mean of the output signal sequence.

[0042] When the potential well parameters of a stochastic resonance (SR) system are mismatched, the output signal sequence Primarily driven by noise, it exhibits a Gaussian-like distribution with a kurtosis value K close to 3.

[0043] When the potential well parameters of a stochastic resonance (SR) system are optimally tuned, a weak fault traveling wave pulse can acquire enough energy to drive particles (system state) to achieve synchronous and violent transitions between the two potential wells, resulting in a sequence of output signals. It exhibits a large amplitude and sharp waveform of impact pulse, with its kurtosis value K being much greater than 3, and reaching its maximum value at this moment.

[0044] Therefore, maximizing the kurtosis value K of the output signal is equivalent to finding the potential well parameter that best enhances the transient impact signal of the stochastic resonance (SR) system. .

[0045] like Figure 2 As shown, the particle swarm optimization (PSO) algorithm in step S3 includes: Step S31, initialize the particle swarm; Set particle swarm size ,For example, .

[0046] Within the preset parameter search space, such as Random initialization The position of each particle and speed ,in Indicates particle index, , This indicates the number of particles participating in the optimization search.

[0047] Set the inertia weights for the Particle Swarm Optimization (PSO) algorithm. Learning factors and maximum number of iterations .

[0048] Step S32, for each particle Substitute the particles into the bistable stochastic resonance model in step S2, and process the preprocessed transient signal. The input is fed into a bistable stochastic resonance model, and numerical solutions are used to obtain the transient signal after preprocessing. output signal Calculate the output signal cliff To output signal cliff As the first The current fitness value of each particle; Step S33: Update the individual optimal and global optimal values ​​based on the particle's fitness value; where individual optimal refers to the best state reached by each particle in its own history, and global optimal refers to the best state reached by the entire particle swarm in its history. Specifically, step S33 includes: updating the current fitness value of each particle... The historical best fitness value of the corresponding particle By comparison, if the particle's fitness value exceeds its historical best fitness value, then... Then the historical best fitness value is updated to the particle's fitness value, that is... And record the corresponding position; at this time, the current position This is its best historical position.

[0049] Fitness values ​​of all particles Compared to the global best fitness value The comparison is performed; if the particle's fitness value exceeds the globally optimal fitness value, then... Then the global best fitness value is updated to the particle's fitness value, that is... And record the corresponding position, at this time the current position The best position globally .

[0050] Step S34, update particle velocity and position; update particle velocity and position using the following formula:

[0051]

[0052] in, The inertial weight is used to balance the global and local search capabilities of a particle. As the number of iterations increases, the inertial weight gradually decreases, allowing the particle to converge to the optimal solution more effectively. This indicates the particle velocity before the update; This indicates the updated particle velocity; , These represent learning factors; , These represent random numbers between [0, 1]. Indicates the optimal position for an individual; Indicates the particle position before the update; Indicates the globally optimal position; Indicates the updated particle position, subscript i Indicates particle index, d This represents the dimension of the potential well parameters, i.e., the parameters mentioned above. or , t、t +1 represents the number of iterations.

[0053] Step S35: Repeat steps S32-S34 until the particle swarm converges or the preset number of iterations is reached, and output the optimized potential well parameters.

[0054] Specifically, the determination is made based on whether the termination condition has been met. The termination condition is that the particle swarm converges, i.e., the global optimal fitness value. If there is no significant change over multiple consecutive generations, or if the preset number of iterations is reached, i.e., the maximum number of iterations mentioned above. The iteration ends when any termination condition is met.

[0055] The globally optimal position obtained after the iteration is completed. That is, the optimized potential well parameters .

[0056] Step S4: Substitute the optimized potential well parameters into the bistable stochastic resonance model to process the preprocessed transient signal and obtain the enhanced output signal; specifically, use the optimized potential well parameters obtained in step S3. Configure a stochastic resonance (SR) system.

[0057] Then, the transient signal stream that has been acquired and preprocessed in real time will be... The signal is continuously input into a well-configured stochastic resonance (SR) system to obtain an enhanced output signal. .

[0058] Step S5: Perform fault detection based on the enhanced output signal to determine whether a fault has occurred.

[0059] Under the condition that the stochastic resonance (SR) system is at its optimal well parameters, when the equipment is in a fault-free operating state, the enhanced output signal... It fluctuates slightly within a certain potential well range. The potential well range referred to here can be either the interval corresponding to a positive value or the interval corresponding to a negative value.

[0060] When a weak fault traveling wave signal arrives, it will trigger one or more clear and significant transitions in the system state, i.e., the enhanced output signal. From a steady state (e.g.) Jump to another stable state (e.g.) ),in .

[0061] This invention, based on the characteristics of system state changes during fault occurrence, transforms the originally complex fault detection problem into a simple threshold comparison problem. Fault assessment is achieved through the threshold comparison method. Step S5 includes the following steps: Based on the optimized potential well parameters, a preset threshold is set; for example, the preset threshold is set using the following formula:

[0062] in, Indicates the safety factor. <1, such as ; , These represent the optimized potential well parameters; This indicates a preset threshold.

[0063] For the enhanced output signal With preset threshold Compare; When the absolute value of the enhanced output signal Exceeding the preset threshold When a fault occurs, the timestamp information of the fault occurrence is recorded.

[0064] Specifically, in the actual fault detection process, transient signals on the cable line are continuously monitored and fed into an optimized bistable stochastic resonance model to obtain an enhanced output signal. When the absolute value of the enhanced output signal is detected. First crossing threshold When the fault occurs, the timestamp of that moment is recorded as the arrival time of the fault traveling wave. This time information can be directly used to trigger subsequent fault location devices, thereby quickly and accurately determining the specific location of the fault.

[0065] This invention utilizes the random resonance (SR) mechanism to transfer some noise energy to weak fault signals, thereby amplifying and enhancing the weak fault signals.

[0066] This invention introduces an adaptive optimization strategy based on maximizing the kurtosis of the output signal. Without human intervention or prior knowledge of the signal, the stochastic resonance (SR) system can automatically find the optimal parameters, which is suitable for detecting non-periodic, transient impulsive traveling wave signals.

[0067] Based on the aforementioned system characteristics, the problem of weak amplitude detection is transformed into a clear and reliable problem of system state transition detection. Under fault-free conditions, the system is in a stable state with minimal output signal fluctuations; when a fault occurs, the system state exhibits a significant transition. By detecting this transition phenomenon, the sensitivity and accuracy of detection can be greatly improved, the false negative rate reduced, and a reliable triggering basis provided for subsequent fault location. This enables the fault location device to be activated promptly and accurately, laying the foundation for rapid fault elimination.

[0068] The following detailed description of the working process of the present invention is illustrated through a specific application scenario.

[0069] Application Scenario: A 5km long 10kV XLPE cable line, with a high-frequency current transformer and signal acquisition module installed at the beginning of the cable line. The sampling frequency of the signal acquisition module... f s =20MHz.

[0070] Suppose that at a specific moment, a high-impedance ground fault occurs 3km from the beginning of a cable line. The weak traveling wave signal generated by the fault will propagate along the line towards the beginning. The specific steps for fault detection are as follows: First, signal acquisition and preprocessing are performed. The signal acquisition module collects the current signal from the cable line in real time. When a slight fluctuation in signal energy is detected, a transient signal containing a data window of 1024 sampling points is captured. Because the amplitude of the fault traveling wave signal is extremely small, it is completely submerged in the strong background noise and cannot be distinguished by the naked eye.

[0071] To facilitate subsequent processing and analysis, the transient signals within this data window are... Perform normalization preprocessing to transform transient signals The amplitude is adjusted to the range [-1, 1] to obtain the preprocessed transient signal. .

[0072] Next, the particle swarm optimization (PSO) algorithm based on kurtosis optimization is launched to adaptively optimize the parameters. First, the particle swarm is initialized, including the particle swarm size. Parameter search range , Inertial weight The learning factor decreased linearly from 0.9 to 0.4. Maximum number of iterations The iterative optimization process is as follows: The algorithm randomly generates 20 particles, such as ; For particles , its parameters and preprocessed transient signals Substituting into a bistable stochastic resonance model, a fourth-order Runge-Kutta algorithm is used, with a step size of... The output signal of the particle was calculated. ; Calculate the output signal kurtosis value .

[0073] Repeat this process for all 20 particles to obtain their respective kurtosis values.

[0074] Update the individual optimality of each particle and the global optimality of the population based on the kurtosis value.

[0075] Update the velocity and position of all particles according to the particle swarm optimization formula.

[0076] After repeating the iteration 50 times, assuming the algorithm converges, the globally optimal parameters are obtained. The corresponding maximum kurtosis value at this time is .

[0077] Finally, signal enhancement and fault diagnosis are performed. First, optimal parameters are used. Construct the final stochastic resonance system; then, process the preprocessed transient signal. The optimal stochastic resonance system is input again to obtain the enhanced output signal. .

[0078] In the original signal, the weak traveling wave pulses are obscured by noise and difficult to identify. However, in the enhanced output signal... In this process, these weak traveling wave pulses are transformed into a dramatic transition from a negative steady state to a positive steady state. The impulse characteristics of the signal are greatly enhanced, while background noise is effectively suppressed to the vicinity of the two steady states.

[0079] In the fault determination process, the criterion threshold is first calculated. Then, continuously monitor the enhanced output signal. When the absolute value of the enhanced output signal is found When the impedance first exceeds 0.67, a high-impedance grounding fault is identified, and the corresponding time is recorded. This trigger signal is then sent to the fault location system.

[0080] Through the above embodiments, the present invention successfully transforms the problem of detecting an extremely weak signal with a possible negative signal-to-noise ratio into a threshold judgment problem of a bistable signal with a high signal-to-noise ratio and obvious characteristics, thereby realizing reliable and sensitive detection of high-impedance grounding faults.

[0081] Compared with the prior art, the present invention has the following significant advantages: (1) This invention utilizes the principle of random resonance to transform the inherent noise in the environment from an interference factor into a beneficial energy source. It can effectively amplify fault signals that are submerged in strong noise and have extremely low amplitude, and the detection sensitivity is much higher than that of traditional methods.

[0082] (2) By introducing a particle swarm optimization (PSO) adaptive optimization mechanism based on kurtosis maximization, this invention can automatically find the optimal operating parameters without any prior knowledge about the signal and noise, such as the signal frequency, amplitude, statistical distribution, etc. It has good adaptability and robustness to high-resistance grounding faults under different lines, different operating conditions, and different noise levels.

[0083] (3) The algorithms involved in this invention, such as the particle swarm optimization (PSO) algorithm and the fourth-order Runge-Kutta algorithm, are mature algorithms with moderate computational complexity. They can be implemented on general-purpose digital signal processors (DSPs), field-programmable gate arrays (FPGAs), or industrial computers. They have low hardware costs, are easy to upgrade or integrate into existing monitoring equipment, and have high engineering application value.

[0084] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made using the content of this specification and illustrations should be included within the protection scope of the present invention.

Claims

1. An adaptive stochastic resonance weak fault detection method based on kurtosis optimization, characterized in that, include: Step S1: Acquire transient signals on the cable line and preprocess the transient signals to obtain preprocessed transient signals; Step S2: Construct a bistable stochastic resonance model based on the Langevin equation, and numerically solve the Langevin equation using the fourth-order Runge-Kutta algorithm to obtain the output signal; Step S3: Use the kurtosis of the output signal as the fitness function, and optimize the potential well parameters of the bistable stochastic resonance model using the particle swarm optimization algorithm to obtain the optimized potential well parameters; Step S4: Substitute the optimized potential well parameters into the bistable stochastic resonance model, process the preprocessed transient signal, and obtain the enhanced output signal; Step S5: Perform fault detection based on the enhanced output signal to determine whether a fault has occurred.

2. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 1, characterized in that, In step S1, the preprocessing includes normalization preprocessing, which uses the following formula: ； in, Indicates a transient signal; n Represents discrete time points; This represents the minimum value of the transient signal within the sliding data window; This represents the maximum value of the transient signal within the sliding data window; This represents the preprocessed transient signal.

3. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 1, characterized in that, In step S2, the differential form of the Langevin equation is: ； in, x This represents the output signal of a random resonance system; a , b ... This represents the preprocessed transient signal; This represents the differential form of the Langevin equation.

4. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 1, characterized in that, In step S2, the output signal is: ； in, , These represent the output signals, n、n +1 represents a discrete time point; a , b ... This represents the preprocessed transient signal; h This indicates the calculation step size.

5. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 1, characterized in that, Step S3 includes: Step S31, initializing the particle swarm; Step S32, for each particle, substituting the particle into the bistable stochastic resonance model to obtain the output signal obtained by numerically solving the preprocessed transient signal, and using the kurtosis of the output signal as the fitness value of the particle; Step S33, updating the individual optimum and the global optimum based on the fitness value of the particle; Step S34, updating the particle velocity and position; Step S35, repeating steps S32-S34 until the particle swarm converges or reaches the preset number of iterations, and outputting the optimized potential well parameters.

6. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 5, characterized in that, Step S33 includes: comparing the fitness value of the particle with the historical best fitness value of the corresponding particle; if the fitness value of the particle exceeds the historical best fitness value, then updating the historical best fitness value to the fitness value of the particle and recording the corresponding position; comparing the fitness value of the particle with the global best fitness value; if the fitness value of the particle exceeds the global best fitness value, then updating the global best fitness value to the fitness value of the particle and recording the corresponding position.

7. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 5, characterized in that, In step S34, the particle velocity and position are updated using the following formula: ； ； in, Indicates inertia weight; This indicates the particle velocity before the update; This indicates the updated particle velocity; , These represent learning factors; , These represent random numbers between [0, 1]. Indicates the optimal position for an individual; Indicates the particle position before the update; Indicates the globally optimal position; Indicates the updated particle position, subscript i Indicates particle index, d Indicates the dimension of the potential well parameters. t、t +1 represents the number of iterations.

8. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 1, characterized in that, Step S5 includes: comparing the enhanced output signal with a preset threshold; when the absolute value of the enhanced output signal exceeds the preset threshold, determining that a fault has occurred, and recording the fault occurrence timestamp information.

9. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 8, characterized in that, Before comparing the enhanced output signal with the preset threshold, the method further includes setting the preset threshold based on the optimized potential well parameters.

10. The adaptive stochastic resonance weak fault detection method based on kurtosis optimization according to claim 8, characterized in that, The preset threshold is: ； in, Indicates the safety factor; , These represent the optimized potential well parameters; This indicates a preset threshold.

Images