An exponential segmented DC arc noise model and its parameter extraction method

Abstract

The present invention discloses an exponential segmented DC arc noise model and a parameter extraction method thereof. The steps of extracting the parameters of the exponential segmented DC arc noise model include: 1) acquiring arc current signals based on current sensors; 2) calculating the spectrum distribution of current signals; 3) fitting the spectrum distribution of current signals using the exponential segmented DC arc noise model; 4) selecting arc noise model parameters to be extracted and constructing an objective function; 5) extracting the parameters of the exponential segmented arc noise model using a feedback chaos growth optimization algorithm. The present invention uses an exponential segmented DC arc noise model to accurately fit the complex nonlinearity of the DC arc noise spectrum under different working conditions. The present invention simultaneously uses a feedback operator and a chaos parameter generation mechanism to improve the growth optimization algorithm, which can improve the evolutionary efficiency of the algorithm and weaken the premature phenomenon in the algorithm convergence process, and can realize the accurate extraction of the parameters of the exponential segmented arc noise model.

Description

Technical Field

[0001] The invention relates to an exponential segmented direct current arc noise model and a parameter extraction method thereof, and belongs to the field of arc fault detection. Technical Background

[0002] In order to cope with energy shortages and environmental pollution, replacing traditional fossil energy with renewable energy is an urgent problem that countries around the world need to solve. Among renewable energy sources, photovoltaic energy has the advantages of cleanliness, high abundance, and low geographical restrictions. In 2021, the global photovoltaic power generation installed capacity will increase by 180GW. Photovoltaic energy is considered to be one of the most promising alternative energy sources. However, during the long-term outdoor operation of the photovoltaic system, due to the effects of external stresses such as weather, animal gnawing, and high working voltage, the internal cables and terminals will deteriorate, which is prone to DC arc faults and fire accidents. Compared with AC arc faults, DC arc faults lack zero crossing points and are difficult to extinguish by themselves. Therefore, DC arc faults are more likely to produce continuous arcing. Standard UL1699B requires that photovoltaic systems with an operating voltage higher than 80V should have arc fault detection functions. In photovoltaic systems, DC arc faults are divided into series arc faults and parallel arc faults. Parallel arc faults trigger circuit overcurrent, and traditional protection devices can easily detect parallel arc faults. A series arc fault is equivalent to introducing a randomly fluctuating impedance into the line, which can cause a slight drop in current. It is difficult for traditional protection devices to detect series arc faults. Therefore, series arc faults are more harmful to photovoltaic systems than parallel arc faults. Series arc faults are highly random and hidden, and how to effectively detect series arc faults in photovoltaic systems is a challenging problem. Therefore, this application focuses on solving the problem of arc faults in DC systems.

[0003] Some scholars have studied arc fault detection methods based on perfect experimental data. However, in actual photovoltaic systems, collecting perfect DC arc fault experimental data at different locations and working conditions is costly, and the generation of DC arc faults is dangerous. Arc models can provide a more flexible and lower-cost means for studying arc fault detection methods, and model-based methods have become an important aspect of arc fault detection method research.

[0004] DC arc fault is a chaotic phenomenon with strong randomness. Arc modeling methods are usually divided into two categories: physical models and black box models. The physical model based on finite element simulation can accurately describe the microscopic physical mechanism of arc plasma discharge, but the physical model is too complex to be applied to the research of arc fault detection methods. The black box model regards the arc as an element in the circuit and is mainly used to describe the external characteristics of the arc without paying attention to its complex physical mechanism. Therefore, this application focuses on the black box model. The polynomial model can simulate the dynamic characteristics of arc noise, but the operation of the polynomial model requires additional experimental data. For example, if a polynomial model for calculating arc current is established, the arc current needs to be calculated based on the actual voltage data. Therefore, the polynomial model cannot completely get rid of the dependence on experimental data in practical applications. Some researchers fit arc noise based on the white noise model. Arc noise exists in the form of colored noise, and the white noise model cannot simulate the nonlinear characteristics of the spectrum. Some researchers use the pink noise model to simulate arc noise. The spectrum energy of the pink noise model is positively correlated with the negative first power of the frequency value. However, the fixed form of the pink noise model cannot accurately fit the various nonlinear characteristics of the arc spectrum. The segmented noise model is an improved form of the pink noise model. Compared with the pink noise model, the segmented noise model can provide more accurate fitting performance. However, the segmented noise model is difficult to accurately describe the strong nonlinearity of the arc spectrum shape through a single decline rate value. Therefore, it is necessary to further improve the segmented noise model so that the decline rate of the spectrum energy shows a nonlinear change with the increase of the frequency value.

[0005] The accuracy of parameter extraction has an important impact on the performance of the arc model. Grid search and least squares methods have been applied to arc model parameter extraction, but both methods have the problems of low efficiency and insufficient parameter extraction accuracy. Metaheuristic algorithms do not need to rely on the mathematical mechanism of the problem to be solved. They have the advantages of simple structure and insensitivity to initial values, and can effectively solve the problem of arc model parameter extraction. Some scholars use genetic algorithms, simulated annealing algorithms, particle swarm algorithms and cuckoo search to extract arc model parameters, but these algorithms have low search efficiency and are prone to fall into local optimality during the iteration process. Therefore, in order to extract arc fault model parameters more accurately, it is necessary to further study new optimization algorithms. Summary of the invention

[0006] In order to solve the problems existing in arc fault modeling and parameter extraction, this invention proposes an exponential segmented DC arc noise model and a feedback chaos growth optimization algorithm.

[0007] In order to achieve the above-mentioned invention object, the present invention specifically adopts the following technical solutions:

[0008] The present invention first provides an exponential segmented DC arc noise model. The mathematical expression of the exponential segmented DC arc noise model is shown in formula (1):

[0009]

[0010] Where f represents the frequency value, f0 represents the frequency turning point, and L can be used to control the energy amplitude of the frequency. w (f) represent the model output spectrum curve and the model input reference spectrum curve respectively. As the frequency value increases, the speed of spectrum energy reduction can be controlled based on parameters k1, k2 and k3. In the present invention, the parameters of the arc model include f0, L, k1, k2 and k3.

[0011] The present invention also provides a parameter extraction method for an exponential segmented DC arc noise model, comprising the following steps:

[0012] Step 1): using a current sensor to collect DC arc current data, and obtaining the spectrum distribution of the arc current based on Fourier transform;

[0013] Step 2): Based on the analysis of the arc current spectrum distribution properties, an exponential segmented DC arc noise model is constructed. The exponential segmented DC arc noise model is shown in formula (1);

[0014]

[0015] Where f represents the frequency value, f0 represents the frequency turning point, and L can be used to control the energy amplitude of the frequency. w (f) represent the model output spectrum curve and the model input reference spectrum curve respectively. As the frequency value increases, the speed of spectrum energy reduction can be controlled based on parameters k1, k2 and k3. In the present invention, the parameters of the arc model include f0, L, k1, k2 and k3.

[0016] Step 3): Design the objective function and determine the arc model parameters to be extracted;

[0017] Step 4): Optimize the parameters to be extracted from the arc model based on the optimization algorithm.

[0018] The present invention also provides a feedback chaos growth optimization algorithm, which improves the reflection mechanism in the original growth algorithm based on the feedback operator. The feedback operator can not only enhance the diversity of the population, but also the historical information in the feedback operator is conducive to accelerating the convergence speed. And based on the logistic chaos map generation algorithm, the random parameters in the algorithm are conducive to the algorithm to fairly call different evolutionary mechanisms, thereby avoiding falling into the local optimum.

[0019] Preferably, the feedback chaos growth optimization algorithm specifically includes the following steps:

[0020] Step 4-1): Set the maximum number of iterations, the population size of the optimization algorithm, and randomly generate the positions of individuals in the population of the optimization algorithm;

[0021] Step 4-2): Calculate the fitness value of each individual in the population of the optimization algorithm according to the objective function;

[0022] Step 4-3): Use the learning mechanism to update the position information of different individuals in the population of the optimization algorithm to simulate the process of students learning in school.

[0023] In the learning mechanism, individuals can identify the differences between different individuals. These differences can be modeled by formula (2).

[0024]

[0025] Among them, Gap k (k = 1, 2, 3 and 4) represents the difference between two individuals (kth group). best is the best individual in the population of the optimization algorithm. better is an individual randomly selected from the next 4 suboptimal individuals. worse is a randomly selected individual from the 5 lowest ranked individuals in the population. L1 and x L2 are two individuals randomly selected from the population.

[0026] Learning Factor (LF k ) is used to indicate the difference in learning effect. k It is described by formula (3).

[0027]

[0028] In the formula, LF k For Gap k The normalized Euclidean norm of , LF k The higher the value, the more the individual learns from the kth group. In a population, different individuals have different abilities. Individuals can use SF i To assess their acceptable range of knowledge. i Calculated by formula (4).

[0029]

[0030] In the formula, GR i is the fitness value of individual i, GR max is the maximum fitness value in the population of the optimization algorithm. i The larger the value, the better the individual is and needs to learn from other individuals. Using the gap between different individuals, the learning mechanism of individual i can be expressed as

[0031]

[0032] Among them, It represents the current iteration number. and are the positions of individual i in the current iteration and the next iteration respectively. The new position of individual i can be calculated by formula (5), but it is necessary to check whether the new position is better than the previous position. In the learning stage, the position of individual i is calculated by formula (6).

[0033]

[0034] Where r1 is a random number in the range of (0,1), and the logistic chaotic map is used to generate the parameter r1, P1 = 0.001. When the individual does not perform well in the new position, P1 can decide whether to accept the new position. fit(·) is the objective function, from which the fitness value of the individual can be obtained. The smaller the fitness value, the better the individual. The learning stage adopts a fusion evolution mechanism, which can accelerate the convergence speed by integrating various information.

[0035] Step 4-4): Use the feedback operator to update the position information of different individuals in the optimization algorithm population to simulate the process of students reflecting on their own learning methods and improving themselves. The feedback operator is shown in formula (7):

[0036]

[0037]

[0038] Where r2 and r3 are random numbers between (0,1) generated by logistic chaotic mapping. P2 = 0.3, which can control the probability of mutation. R is an individual randomly selected from the five individuals with the smallest fitness value in the population. is an individual randomly selected from the population of the previous generation optimization algorithm, is a randomly selected individual in the population of this iteration. rstore(·) is a random sorting operator. The position of individual i is calculated by formula (6). r4 and r5 are random numbers generated by standard normal distribution according to formula (9).

[0039]

[0040] In the formula, θ1 and θ2 are generated by the logistic chaotic map generation parameters and are in the range of (0,1). The expression of the logistic chaotic map is shown in formula (10):

[0041] ρ It+1 =cρ It (1-ρ It ) (10)

[0042] Where c = 0.4, 0 < ρ It <1. When the algorithm is executed, r1, r2, r3, θ1 and θ2 are used to replace the variable ρ in formula (10) to generate chaotic sequences corresponding to the parameters r1, r2, r3, θ1 and θ2 respectively.

[0043] Step 4-5): Calculate the current number of iterations. If the current number of iterations is greater than the maximum number of iterations, go to step 4-7), otherwise go to step 4-6);

[0044] Step 4-6): Increase the number of iterations by 1 and go to step 4-2);

[0045] Step 4-7): The result of the parameter extraction of the exponential segmented DC arc noise model is the position vector of the optimal individual in the population of the optimization algorithm.

[0046] The beneficial effects of the present invention are:

[0047] 1. The present invention proposes an exponential segmented DC arc noise model, which uses an exponential function to improve the parameters of the traditional segmented noise model that control the frequency drop rate. Compared with the traditional segmented DC arc noise model, the output of the exponential segmented DC arc noise model can more effectively fit the phenomenon that the energy value drop rate of the actual arc current spectrum is inconsistent in the low frequency band and the high frequency band.

[0048] 2. The present invention proposes a feedback chaos growth optimization algorithm by introducing feedback operators and chaos mechanisms into the growth algorithm. In the feedback chaos growth optimization algorithm, the feedback operator is used to replace the reflection mechanism in the original growth algorithm, which can enhance the probability of the population evolving in the optimal direction, and at the same time, the historical information can be used to improve the convergence speed. The key random parameters are generated by chaotic mapping, which can effectively prevent the population from falling into the local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart of realizing arc model parameter optimization based on feedback chaotic growth optimization algorithm according to one embodiment of the present invention;

[0050] Figure 2 The arc noise spectrum output by the exponential segmented DC arc noise model of an embodiment of the present invention and the actual arc current noise spectrum;

[0051] Figure 3 is an arc noise spectrum output by a pink noise model of an embodiment of the present invention and an actual arc current noise spectrum;

[0052] Figure 4 The arc noise spectrum output by the exponential piecewise DC noise model of an embodiment of the present invention and the actual arc current noise spectrum; DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] like Figure 1 As shown, this embodiment provides an exponential segmented DC arc noise model and a parameter extraction method thereof, comprising the following steps:

[0055] Step 1): The arc current signal is collected at a frequency of 200 kHz based on a current sensor, and the frequency spectrum of the current signal is obtained using a fast Fourier transform method.

[0056] Step 2): Construct an exponential segmented DC arc noise model based on the distribution curve characteristics of the arc current spectrum, as shown in formula (1):

[0057]

[0058] Where f and f0 represent the frequency value and the frequency turning point respectively, and L can be used to control the energy amplitude of the frequency. w (f) represent the model output spectrum curve and the model input reference spectrum curve, respectively, where S w The original signal of (f) conforms to the form of normal distribution. As the frequency value increases, the speed of spectrum energy reduction can be controlled based on parameters k1, k2 and k3. In the present invention, the parameters of the arc model include f0, L, k1, k2 and k3.

[0059] Step 3): Construct the objective function based on the form of root mean square error:

[0060]

[0061] Here, NE represents the number of data points. In this embodiment, the NE value is set to 20, that is, the frequency domain of 0 to 100 kHz is evenly divided into 20 segments.

[0062] e(·) is the error equation, and its expression is shown in formula (12):

[0063]

[0064] In the formula, is the sum of the spectrum energy of the arc current in frequency band i. i Po is the sum of the spectrum energy of the analog data in frequency band i. iRefers to how many data points are contained in frequency band i.

[0065] Step 4): Optimize the parameters extracted from the exponential segmented DC arc noise model based on the optimization algorithm. The present invention adopts a swarm intelligence optimization algorithm. By distributing a group containing a certain number of individuals in the search space and making the individuals evolve according to a certain mechanism, the optimal value in the search space is gradually approached to achieve the purpose of optimization.

[0066] The optimization algorithm is a feedback chaotic growth optimization algorithm. The reflection mechanism in the original growth algorithm is improved based on the feedback operator. The feedback operator can not only enhance the diversity of the optimization algorithm's population, but also the historical information in the feedback operator is conducive to accelerating the convergence speed. The random parameters in the algorithm are generated based on the logistic chaotic map, which is conducive to the algorithm to fairly call different evolutionary mechanisms, thereby avoiding falling into local optimality.

[0067] Step 4-1): Set the maximum number of iterations to 500, the population size in the optimization algorithm to 20, the parameters P1 and P2 to 0.001 and 0.3 respectively, and randomly generate the positions of individuals in the population;

[0068] Step 4-2): Calculate the fitness value of each individual in the population in the optimization algorithm according to the objective function;

[0069] Step 4-3): Use the learning mechanism to update the position information of different individuals in the population in the optimization algorithm to simulate the process of students learning in school.

[0070] In the learning mechanism, individuals can identify the gaps between different individuals. These gaps can be modeled by formula (4).

[0071]

[0072] Among them, Gap k (k = 1, 2, 3 and 4) represents the difference between two individuals (kth group). best is the best individual in the population of the optimization algorithm. better is an individual randomly selected from the next 4 suboptimal individuals. worse is a randomly selected individual from the 5 lowest ranked individuals in the population. L1 and x L2 are two individuals randomly selected from the population of the optimization algorithm.

[0073] Learning Factor (LF k ) is used to indicate the difference in learning effect. k It is described by formula (5).

[0074]

[0075] In the formula, LF k For Gap k The normalized Euclidean norm of , LF k The higher the value, the more the individual learns from the kth group. In a population, different individuals have different abilities. Individuals can use SF i To assess their acceptable range of knowledge. i Calculated by formula (6).

[0076]

[0077] In the formula, GR i is the fitness value of individual i, GR max is the maximum fitness value in the population of the optimization algorithm. i The larger the value, the better the individual is and needs to learn from other individuals. Using the gap between different individuals, the learning mechanism of individual i can be expressed as

[0078]

[0079] Among them, It represents the current iteration number. and are the positions of individual i in the current iteration and the next iteration respectively. The new position of individual i can be calculated by formula (7), but it is necessary to check whether the new position is better than the previous position. In the learning stage, the position of individual i is calculated by formula (8).

[0080]

[0081] Where r1 is a random number in the range of (0,1), and the logistic chaotic map is used to generate the parameter r1. When the individual does not perform well in the new position, P1 can decide whether to accept the new position. fit(·) is the objective function, from which the fitness value of the individual can be obtained. The smaller the fitness value, the better the individual. The learning stage adopts a fusion evolution mechanism, which can accelerate the convergence speed by integrating various information.

[0082] Step 4-4): Use the feedback operator to update the position information of different individuals in the optimization algorithm population to simulate the process of students reflecting on their own learning methods and improving themselves. The feedback operator is shown in formula (9):

[0083]

[0084]

[0085] Where r2 and r3 are random numbers between (0,1) generated by logistic chaotic mapping. P2 can control the probability of mutation. R is an individual randomly selected from the five individuals with the smallest fitness value in the population of the optimization algorithm. is an individual randomly selected from the previous generation population. rstort(·) is a random sorting operator. The position of individual i is calculated by formula (9). r4 and r5 are random numbers generated by standard normal distribution by formula (11).

[0086]

[0087] Where θ1 and θ2 are generated by the logistic chaotic map generation parameters and are in the range of (0,1).

[0088] The expression of logistic chaotic mapping is shown in formula (12):

[0089] ρ It+1 =cρ It (1-ρ It ) (12)

[0090] Where c = 0.4, 0 < ρ It <1, when the algorithm is executed, r1, r2, r3, θ1 and θ2 are used to replace the variable ρ to generate the chaotic sequences corresponding to r1, r2, r3, θ1 and θ2 respectively.

[0091] Step 4-5): Calculate the current number of iterations. If the current number of iterations is greater than the maximum number of iterations, go to step 4-7), otherwise go to step 4-6);

[0092] Step 4-6): Increase the number of iterations by 1 and go to step 4-2);

[0093] Step 4-7): The result of the parameter extraction of the exponential segmented DC arc noise model is the position vector of the optimal individual in the population of the optimization algorithm.

[0094] The above embodiment can realize accurate identification of the parameters of the exponential segmented DC noise model. The arc noise spectrum output by the exponential segmented DC arc noise model is similar to the actual arc current noise spectrum. Figure 2 The spectrum of the model output data can accurately fit the complex nonlinearity of the actual current noise signal spectrum, and the root mean square error calculated by formula (2) is 0.0230 (as shown in Table 1).

[0095] Table 1 Results of exponential segmented DC arc noise model parameters identified based on different optimization algorithms

[0096]

[0097] The root mean square errors of model parameter identification based on feedback chaos growth optimization algorithm, growth optimization algorithm, Kepler optimization algorithm, spider wasp optimizer, ant lion optimizer and augmented gray wolf optimization are 0.0230, 0.0247, 0.0344, 0.0298, 0.0405 and 0.0313 respectively. The parameter identification accuracy of feedback chaos growth optimization algorithm has better convergence accuracy than the other five comparison algorithms. The use of feedback operator in the feedback chaos growth optimization algorithm of the present invention can improve the orderliness of population evolution and effectively reduce the precocious phenomenon. Chaotic mapping has stronger ergodicity than uniform distribution. Generating key random parameters in FCGO based on chaotic mapping is conducive to fair use of different evolutionary mechanisms, so that the algorithm maintains a balance between exploration and utilization. Experimental results prove the advancement of the feedback chaos growth optimization algorithm proposed by the present invention.

[0098] The arc noise spectrum output by the pink noise model and the actual arc current noise spectrum are shown in the figure below: Figure 3 The arc noise spectrum output by the segmented DC noise model and the actual arc current noise spectrum are shown in Figure 4 shown.

[0099] pass Figure 2 and Figure 3 It can be seen from the comparison that in the frequency bands of [7kHz, 20kHz] and [50kHz, 100kHz], the proposed exponential segmented DC noise model has a significantly better fitting accuracy for the actual arc current noise spectrum than the pink noise model. This is because the energy of the pink noise model is positively correlated with the negative first power of the frequency value in the entire frequency band, while the spectrum of the arc noise shows an obvious segmented characteristic in the entire frequency band.

[0100] pass Figure 2 and Figure 4 From the comparison, it can be seen that in the [1kHz, 3kHz] frequency band, the spectrum energy output by the piecewise DC noise model is significantly higher than the spectrum energy corresponding to the actual data, while the spectrum energy output by the exponential piecewise DC noise model effectively overlaps with the spectrum energy corresponding to the actual data. In other frequency bands, the performance difference between the exponential piecewise DC noise model and the piecewise DC noise model is difficult to intuitively display.

[0101] The embodiment calculates the root mean square error (modeling error) obtained when different models fit the actual arc current data spectrum. The smaller the root mean square error, the higher the model accuracy. The root mean square error of the traditional pink noise model is 0.1478, and the root mean square error of the segmented DC arc noise model is 0.0667. The root mean square error of the exponential segmented DC noise model proposed in the present invention is only 0.0230. The root mean square error of the pink noise model and the root mean square error of the segmented DC arc noise model are 6.42 times and 2.9 times the root mean square error of the exponential segmented DC noise model, respectively. Moreover, the exponential segmented DC noise model uses an exponential function to make the model output spectrum decrease in a variable power, while the spectrum decrease rate of the segmented DC noise model is a constant, so the exponential segmented DC noise model proposed in the present invention has stronger flexibility than the segmented DC noise model.

[0102] The above experimental results show that compared with the pink noise model and the segmented DC arc noise model, the exponential segmented DC noise model proposed in the present invention can more accurately characterize the complex nonlinear characteristics of the arc noise spectrum.

[0103] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A parameter extraction method for an exponential segmented DC arc noise model, characterized in that: The following steps are involved: Step 1): using a current sensor to collect DC arc current data, and obtaining the spectrum distribution of the arc current based on Fourier transform; Step 2): Based on the analysis of the arc current spectrum distribution properties, an exponential segmented DC arc noise model is constructed. The exponential segmented DC arc noise model is shown in formula (1); The parameters of the arc model include f0, L, k1, k2 and k3, where f represents the frequency value, f0 represents the frequency turning point, L can be used to control the energy amplitude of the frequency, S(f) and S w (f) represent the arc model output spectrum curve and the arc model input reference spectrum curve respectively; as the frequency value increases, the decreasing speed of the spectrum energy is controlled based on parameters k1, k2 and k3; Step 3): Design the objective function and determine the exponential segmented DC arc noise model parameters to be extracted; Step 4): Extract the parameters to be extracted from the exponential segmented DC arc noise model based on the feedback chaos growth optimization algorithm. The feedback chaos growth optimization algorithm introduces feedback operators and chaos mechanisms into the growth optimization algorithm, improves the reflection mechanism in the original growth algorithm based on the feedback operator, and generates random parameters in the algorithm based on the logistic chaos mapping to avoid falling into local optimality.

2. The parameter extraction method of the exponential segmented DC arc noise model according to claim 1, characterized in that: In step 4), the location information of different individuals in the population of the optimization algorithm is updated using a learning mechanism to simulate the process of students learning in school; The feedback operator is used to update the position information of different individuals in the population of the optimization algorithm to simulate the process of students reflecting on their own learning methods and improving themselves. The iteration is performed until the preset maximum number of iterations is reached. At this time, the result of the exponential segmented DC arc noise model parameter extraction is the position vector of the optimal individual in the population of the optimization algorithm.

3. The parameter extraction method of the exponential segmented DC arc noise model according to claim 2, characterized in that: In step 4), in the learning mechanism, individuals in the population of the optimization algorithm can identify the gap between different individuals, and the gap is modeled by formula (2): Among them, Gap k , k = 1, 2, 3, 4, represents the gap between two individuals in the kth group, x best is the individual with the smallest fitness value in the population of the optimization algorithm, that is, the optimal individual, x better is an individual randomly selected from the next several suboptimal individuals, x worse is an individual randomly selected from the lowest ranked individuals in the population; L1 and x L2 are two individuals randomly selected from the population of the optimization algorithm; Learning Factor LF k Used to indicate the difference in learning effect, LF k Described by formula (3): In the formula, LF k For Gap k The normalized Euclidean norm of , LF k The higher the value, the more the individual learns from the kth group; in a population, different individuals have different abilities, and individuals use SF i To assess their acceptable range of knowledge, SF i Calculated by formula (4) In the formula, GR i is the fitness value of individual i, GR max is the maximum fitness value in the population of the optimization algorithm, GR i The larger the value, the less good the individual is and the more it needs to learn from other individuals.

4. The parameter extraction method of the exponential segmented DC arc noise model according to claim 2 or 3, characterized in that: In step 4), in the learning mechanism, the gap between different individuals is used, and the learning mechanism of individual i is expressed as Among them, It represents the current iteration number; and are the positions of a single i in the current iteration and the next iteration respectively; the learning factor LF k Used to indicate the difference in learning effect. Personally, I use SF i To assess their acceptable knowledge range, Gap k , k = 1, 2, 3, 4, indicating the gap between the two individuals in the kth group.

5. The parameter extraction method of the exponential segmented DC arc noise model according to claim 4, characterized in that: In step 4), in the learning mechanism, the new position of individual i is calculated by formula (5), but it is necessary to check whether the new position is better than the previous position; in the learning stage, the position of individual i is calculated by formula (6): Where r1 is a random number in the range of (0,1). When the fitness value of an individual in a new position is not good enough, P1 is a parameter in the optimization algorithm that can decide whether to accept the new position. fit(·) is the objective function, from which the fitness value of the individual can be obtained. The smaller the fitness value, the better the individual. The learning stage adopts a fusion evolution mechanism, which can accelerate the convergence speed by integrating various information.

6. The parameter extraction method of the exponential segmented DC arc noise model according to claim 5, characterized in that: In step 4), the feedback operator is shown in formula (7): Where r2 and r3 are random numbers between (0,1) generated by uniform distribution; P2 is a parameter that controls the probability of mutation, and R is an individual randomly selected from a number of individuals with the smallest fitness value in the population of the optimization algorithm; are individuals randomly selected from the previous generation population; is a randomly selected individual in the population of this iteration; rstore(·) is a random sorting operator; the position of individual i is calculated by formula (6); r4 and r5 are random numbers.

7. The parameter extraction method of the exponential segmented DC arc noise model according to claim 6, characterized in that: The r4 and r5 are random numbers generated by the standard normal distribution according to formula (9): Where θ1 and θ2 are generated by uniform distribution and are in the range of (0,1). Therefore, r4 and r5 are indirectly generated by uniform distribution.

8. The parameter extraction method of the exponential segmented DC arc noise model according to claim 7, characterized in that: In step 4), the parameters r1, r2, r3, θ1 and θ2 are generated based on the logistic chaotic map. The expression of the logistic chaotic map is shown in formula (10): r It+1 =cρ It (1-p It ) (10) Where c = 0.4, 0 < ρ It <1, when the algorithm is executed, r1, r2, r3, θ1 and θ2 are used to replace the variable ρ in formula (10) to generate chaotic sequences corresponding to the parameters r1, r2, r3, θ1 and θ2 respectively.

9. The parameter extraction method of the exponential segmented DC arc noise model according to claim 1, characterized in that: In step 3), the objective function is constructed based on the form of root mean square error: Where NE represents the number of data points, the frequency band of the arc current is evenly divided into NE segments, and e(·) is the error equation, which is expressed as shown in formula (12): In the formula, is the energy sum of the i-th frequency band of the arc current; Sum i Po is the energy sum of the i-th frequency band of the simulated data in FB(i); i Refers to how many data points are contained in frequency band i.

Images