Circuit breaker unknown fault diagnosis method based on IGOA-SVDD

Abstract

The invention discloses a circuit breaker unknown fault diagnosis method based on IGOA-SVDD, and the method comprises the following steps: S01, taking an unknown fault diagnosis problem as an abnormal data detection problem, and carrying out the modeling through employing a support vector data description (SVDD) algorithm; s02, aiming at the problem that a locust algorithm (GOA) is easy to fall into local optimum, improving the locust algorithm by adopting a cloud model inertia weight and a chaos reverse learning algorithm, and providing an improved locust algorithm (IGOA); s03, aiming at the problem that a penalty factor and a kernel function in a support vector data description algorithm are difficult to determine, optimizing the support vector data description algorithm by adopting an improved locust algorithm, and proposing a diagnosis model based on IGOA-SVDD; and S04, applying the diagnosis model based on the IGOA-SVDD to unknown fault diagnosis of the 110kV circuit breaker, and obtaining a diagnosis result. The diagnosis model of the method is applied to unknown fault diagnosis of the 110kV circuit breaker, the unknown fault problem of the 110kV circuit breaker can be effectively solved, and the classification precision is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit breaker fault diagnosis, and in particular to a circuit breaker unknown fault diagnosis method based on IGOA-SVDD. Background Art

[0002] High-voltage circuit breakers are the main power control equipment in substations and have arc extinguishing characteristics. When the system operates normally, they can cut off and connect the no-load and load currents of the lines and various electrical equipment. When a system failure occurs, it cooperates with the relay protection to quickly switch the fault current and prevent the scope of the accident from expanding. Therefore, high-voltage circuit breakers play an important role in substations.

[0003] In actual use, high-voltage circuit breakers may have a variety of faults, and the common fault conditions are as follows:

[0004] In overload conditions, the load current exceeds the rated current of the circuit breaker, causing the circuit breaker to automatically trip to prevent the equipment from overheating, burning or damage;

[0005] In the case of a short circuit, the circuit breaker trips due to an abnormal increase in current caused by a fault in the internal circuit of the equipment;

[0006] ‌ In the case of poor contact, frequent tripping or arcing may occur due to loose or contaminated internal contact parts of the equipment;

[0007] The power supply voltage is too low or too high, the power supply is interrupted, the control circuit power supply fails, etc. will affect the normal operation of the circuit breaker.

[0008] In addition to the above-mentioned fault problems, circuit breakers may also have some unknown faults. At present, when using relevant diagnostic models to diagnose unknown faults of circuit breakers, the training data only includes normal sample data or unknown faults encountered during the test process. It is impossible to diagnose fault types that have never occurred. Summary of the invention

[0009] In order to overcome the above problems, the purpose of the present invention is to provide a circuit breaker unknown fault diagnosis method based on IGOA-SVDD. The method proposes an IGOA-SVDD diagnostic model and applies it to the unknown fault diagnosis of 110kV circuit breakers. It can effectively solve the unknown fault problem of 110kV circuit breakers, has high classification accuracy, and more accurate diagnostic results.

[0010] The technical solution adopted by the present invention is:

[0011] The circuit breaker unknown fault diagnosis method based on IGOA-SVDD includes the following steps:

[0012] S01: The unknown fault diagnosis problem is regarded as an abnormal data detection problem and the support vector data description (SVDD) algorithm is used for modeling;

[0013] S02: Aiming at the problem that the locust algorithm (GOA) is prone to fall into the local optimum, the cloud model inertia weight and chaos reverse learning algorithm are used to improve the locust algorithm, and an improved locust algorithm (IGOA) is proposed;

[0014] S03: Aiming at the problem that the penalty factor and kernel function in the support vector data description algorithm are difficult to determine, the improved locust algorithm is used to optimize the support vector data description algorithm, and a diagnostic model based on IGOA-SVDD is proposed;

[0015] S04: Use the IGOA-SVDD diagnostic model to diagnose unknown faults of 110kV circuit breakers and obtain diagnostic results.

[0016] As a further description of the present invention, the steps of the improved locust algorithm in S02 are as follows:

[0017] S021: Use the chaotic reverse learning strategy to initialize the population so that the population is distributed as evenly as possible in the feasible solution space, improving the uniformity and diversity of the initial population distribution of the locust algorithm;

[0018] S022: Introduce the inertia weight cloud model into the locust algorithm and use different inertia weight strategies to adjust the convergence speed of the algorithm;

[0019] S023: Based on the chaotic logistic mapping principle, local deep search is performed to optimize the later iterations of the locust algorithm, reduce the probability of falling into the local optimum, specify the optimal fitness value, and execute the first two steps again until the end standard is reached.

[0020] As a further description of the present invention, the population initialization based on the Logistic chaotic adversarial learning strategy is adopted in S021 to improve the uniformity and diversity of the initial population distribution of the locust algorithm in two stages, specifically:

[0021] In the first stage, the chaotic sequence is generated by using the logistic mapping of formula (1) to initialize the population:

[0022] Formula (1)

[0023] In the formula,

[0024] represents the chaotic variable,

[0025] represents the sequence number of the chaotic variable,

[0026] represents the sequence number of the population,

[0027] r represents the dimension of the population;

[0028] In the second stage, we use equations (2) and (3) to obtain a new population, calculate and sort the new population, and select the best N individuals as the initial population. The specific calculation process is:

[0029] Simultaneously calculate and evaluate a candidate solution and the corresponding matching inverse solution, select the optimal solution as the next generation individual, and , calculate the opposite solution of the relevant match according to the following formula ,and ( represents the upper boundary, represents the lower boundary), calculated as follows:

[0030]

[0031] Convert the above formula from one dimension to high dimension as follows:

[0032]

[0033] Map the chaotic sequence into the solution space,

[0034] Get population , Formula (2)

[0035] Population individuals , Locust The first The initial value of the dimension, and To search for upper and lower bounds,

[0036] By population Calculate the reverse population

[0037] The reverse population is expressed as: Formula (3)

[0038] The population and Merge to get a new population.

[0039] As a further description of the present invention, the inertia weight of the cloud model is used in S022 to divide the locust population into good subgroups, ordinary subgroups and bad subgroups, and inertia weights of different sizes are used to improve the convergence speed and expand the search range. The modified mathematical model is expressed as follows:

[0040] Formula (4)

[0041] In the formula,

[0042] , Respectively Locust The upper and lower bounds of the dimensional variables,

[0043] is the target location of the locust swarm,

[0044] is a linearly decreasing coefficient.

[0045] As a further description of the present invention, the SVDD model in S01 uses the hypersphere formed by the support vector as the boundary line, the data within the boundary is called the training sample, and the data outside the boundary is called other types of data. The decision function is used to determine whether the test data and the training data are of the same type. When the decision function is 1 and the distance between the test data and the center of the hypersphere is smaller than the radius of the hypersphere, it means that the test data and the training data are of the same type, otherwise they are of other types. The specific definition is as follows:

[0046] Assuming the radius of the hypersphere is R and the center is a, the expression of SVDD is:

[0047] ,

[0048] In the formula,

[0049] , is called the Lagrangian operator, is the support vector,

[0050] When classifying a given sample, calculate the distance between it and the hypersphere , and compared with the radius R of the hypersphere, the distance formula The details are as follows:

[0051] ,

[0052] The above formula can be used to determine whether it is a known working condition or an unknown working condition.

[0053] As a further description of the present invention, the SVDD model uses G-means as the fitness function, and the specific expression is as follows:

[0054] ,

[0055] The calculation formula of G-mean is as follows:

[0056] ,

[0057] In the formula,

[0058] TP represents the number of correct classifications in the positive category.

[0059] FN represents the number of errors in the positive category.

[0060] TN represents the number of correct negative categories.

[0061] FP represents the number of incorrect classifications in the negative category.

[0062] As a further description of the present invention, the steps of establishing the diagnostic model based on IGOA-SVDD in S03 are:

[0063] S031: Initialize algorithm parameters and set population size , maximum number of iterations ;

[0064] S032: Initialize the locust population algorithm using the reverse chaos learning algorithm;

[0065] S033: Use the inertia weight of the cloud model to divide the locust population into three categories: optimal population, suboptimal population and general population, optimize the penalty factor C and kernel parameter δ to be optimized, and calculate the fitness value of the locust individuals;

[0066] S034: Compare the current optimal result with the optimal result of the previous step optimization. If the current result is better than the previous step optimization result, the current fitness value is taken as the optimal value.

[0067] S035: Use formula (4) to calculate the probability of locust individuals to avoid falling into the local optimum;

[0068] S036: Determine whether the final condition is met. If the final condition is met, input the parameters to be optimized into the SVDD model. Otherwise, return to step S032.

[0069] As a further description of the present invention, in the S04, the PSO-VMD decomposition method is used to decompose the vibration signal when diagnosing and analyzing the fault, and the vibration signal is extracted from the time domain, frequency domain, energy and composite multi-scale weighted permutation entropy respectively to form a multi-dimensional feature fusion matrix. The vibration signal represented by N samples is collected for a total of S segments. The feature matrix can be expressed as:

[0070]

[0071] In the formula,

[0072] J=s+k+15 represents the dimension of the feature matrix.

[0073] S represents the number of locust algorithms,

[0074] Considering the length and complexity of data calculation, the mRMR principle is used to screen the vibration signal, and the reduced feature matrix is ​​defined as follows:

[0075]

[0076] As a further description of the present invention, the optimal individual chaotic search strategy is used in S033 to find the optimal solution.

[0077] Beneficial effects of the present invention:

[0078] The invention discloses a circuit breaker unknown fault diagnosis method based on IGOA-SVDD. The IGOA-SVDD-based diagnostic model is used in the unknown fault diagnosis of 110kV circuit breakers, which can effectively solve the unknown fault problem of 110kV circuit breakers, has high classification accuracy, and more accurate diagnostic results. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 This is a flow chart of the circuit breaker unknown fault diagnosis method based on IGOA-SVDD proposed by the present invention;

[0080] Figure 2 This is a specific flow chart of S03 of the circuit breaker unknown fault diagnosis method based on IGOA-SVDD proposed by the present invention;

[0081] Figure 3 It is a schematic diagram of the optimization process of Ackely function by different optimization algorithms in the embodiment of the circuit breaker unknown fault diagnosis method based on IGOA-SVDD proposed by the present invention;

[0082] Figure 4 It is a schematic diagram of the optimization process of the Griewank function by different optimization algorithms in the embodiment of the circuit breaker unknown fault diagnosis method based on IGOA-SVDD proposed by the present invention;

[0083] Figure 5 A comparison diagram of fault diagnosis results of different methods in an embodiment of the circuit breaker unknown fault diagnosis method based on IGOA-SVDD proposed by the present invention;

[0084] Figure 6 This is a comparison chart of different algorithms under different conditions in an embodiment of the circuit breaker unknown fault diagnosis method based on IGOA-SVDD proposed by the present invention. DETAILED DESCRIPTION

[0085] The specific implementation of the present invention is described below in conjunction with the accompanying drawings and embodiments:

[0086] It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical contents disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0087] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0088] like Figures 1 to 6 As shown, it shows a specific embodiment of the present invention:

[0089] Embodiment 1

[0090] The circuit breaker unknown fault diagnosis method based on IGOA-SVDD includes the following steps:

[0091] S01: The unknown fault diagnosis problem is regarded as an abnormal data detection problem and the support vector data description (SVDD) algorithm is used for modeling;

[0092] S02: Aiming at the problem that the locust algorithm (GOA) is prone to fall into the local optimum, the cloud model inertia weight and chaos reverse learning algorithm are used to improve the locust algorithm, and an improved locust algorithm (IGOA) is proposed;

[0093] S03: Aiming at the problem that the penalty factor and kernel function in the support vector data description algorithm are difficult to determine, the improved locust algorithm is used to optimize the support vector data description algorithm, and a diagnostic model based on IGOA-SVDD is proposed;

[0094] S04: Use the IGOA-SVDD diagnostic model to diagnose unknown faults of 110kV circuit breakers and obtain diagnostic results.

[0095] Embodiment 2

[0096] Specifically, the steps of the improved locust algorithm in S02 are as follows:

[0097] S021: Use the chaotic reverse learning strategy to initialize the population so that the population is distributed as evenly as possible in the feasible solution space, thereby improving the uniformity and diversity of the initial population distribution of the locust algorithm;

[0098] S022: Introduce the inertia weight cloud model into the locust algorithm and use different inertia weight strategies to adjust the convergence speed of the algorithm;

[0099] S023: Based on the chaotic logistic mapping principle, local deep search is performed to optimize the later iterations of the locust algorithm, reduce the probability of falling into the local optimum, specify the optimal fitness value, and execute the first two steps again until the end standard is reached.

[0100] In this embodiment, the IGOA is an improved GOA algorithm, which achieves the target optimization goal based on the small-range movement of locust larvae and the large-range search of adults and the overall movement. The mathematical model of the locust algorithm is as follows:

[0101] ,

[0102] In the formula, represents the position of the i-th locust, represents the interaction between the i-th locust and other locusts in the group, Defines the gravity of the i-th locust, Represents the impact of wind on the i-th locust.

[0103] To provide random behavior, the above formula can be changed to:

[0104]

[0105] In the formula, , and represents a random number between [0,1], and , , Affected by different environmental factors, they are expressed as:

[0106] ,

[0107] In the formula, represents the distance between the i-th locust and the j-th locust. , It represents the unit vector from the i-th locust to the j-th locust. The mathematical function s is defined as the interaction effect in the locust population, and the expression is as follows:

[0108] ,

[0109] In the formula, is the attraction constant, is the attraction length scale, , .

[0110] ,

[0111] Where g is the gravitational constant, is a unit vector.

[0112] ,

[0113] In the formula, represents the drift constant, It represents the unit vector of wind direction.

[0114] therefore, , N represents the number of locusts.

[0115] The optimization problem is solved by the mathematical model of the above formula. The locust swarm can quickly reach the comfort zone without converging to the destination. The above formula can also be transformed into:

[0116] ,

[0117] In the formula, and The representative dimension is The upper and lower bounds of the dimension, The dimension is The current search value of . and represents a linearly decreasing coefficient. and The expression is as follows:

[0118] ,

[0119] In the formula, and are the maximum and minimum values, respectively. Indicates the current iteration number, is the maximum number of iterations, , .

[0120] Specifically, the population initialization based on the Logistic chaotic adversarial learning strategy is adopted in S021 to improve the uniformity and diversity of the initial population distribution of the locust algorithm in two stages, specifically:

[0121] In the first stage, the chaotic sequence is generated by using the logistic mapping of formula (1) to initialize the population:

[0122] Formula (1)

[0123] In the formula,

[0124] represents the chaotic variable,

[0125] represents the sequence number of the chaotic variable,

[0126] represents the sequence number of the population,

[0127] r represents the dimension of the population;

[0128] In the second stage, we use equations (2) and (3) to obtain a new population, calculate and sort the new population, and select the best N individuals as the initial population. The specific calculation process is:

[0129] Simultaneously calculate and evaluate a candidate solution and the corresponding matching inverse solution, select the optimal solution as the next generation individual, and , calculate the opposite solution of the relevant match according to the following formula ,and ( represents the upper boundary, represents the lower boundary), calculated as follows:

[0130]

[0131] Convert the above formula from one dimension to high dimension as follows:

[0132]

[0133] Map the chaotic sequence into the solution space,

[0134] Get population , Formula (2)

[0135] Population individuals , Locust The first The initial value of the dimension, and To search for upper and lower bounds,

[0136] By population Calculate the reverse population

[0137] The reverse population is expressed as: Formula (3)

[0138] The population and Merge to get a new population.

[0139] Specifically, in S022, the inertia weight of the cloud model is used to divide the locust population into good subgroups, ordinary subgroups and bad subgroups, and inertia weights of different sizes are used to improve the convergence speed and expand the search range. The corrected mathematical model is expressed as follows:

[0140] Formula (4)

[0141] In the formula,

[0142] , Respectively Locust The upper and lower bounds of the dimensional variables,

[0143] is the target location of the locust swarm,

[0144] is a linearly decreasing coefficient.

[0145] In this embodiment, in the locust algorithm, the inertia weight Very important for local and global development, large inertia weight It is helpful to jump out of the local optimum, but it is not easy to get an exact solution, and the inertia weight is small. It is conducive to local optimality and accelerates the convergence speed of the algorithm, but it is not easy to jump out of the local extreme point. Therefore, first calculate the fitness value of all locusts and find the average fitness value of all locusts. Assume that the total number of locusts is , in In the iterations, a single The fitness value is The corresponding average fitness value is .

[0146] When the individual fitness value is greater than When the average fitness value is ; When the individual fitness value is less than When the average fitness value is ; The best individual fitness value is Based on the above classification, the locust population is divided into three subgroups with different inertia weight generation strategies:

[0147] if , then the value is very small, close to the optimal solution, and The value is 0.2;

[0148] if , then it is a general locust population, and the inertia weight can be expressed by Conditional cloud generator adaptive adjustment, particle application The formula for the conditional cloud generator to dynamically and adaptively adjust the inertia weight is as follows:

[0149] ,

[0150] In the formula, and is the learning factor, expected value Points representing qualitative concepts, entropy It is a measure of the degree of fuzziness of qualitative concepts, super entropy It is a measure of the randomness of the membership function, and its size indirectly reflects the thickness of the cloud. The calculation formula is as follows:

[0151] ,

[0152] ,

[0153] According to the mathematical limit theorem:

[0154] The weight range can ensure that ,

[0155] therefore, will decrease as the individual fitness value of the locust decreases, thus ensuring that the better locusts can obtain smaller If , locusts are the worst individuals far from the optimal solution, so the inertia weights of these locusts can be increased, making .

[0156] In this embodiment, in the later stage of the iteration of the GOA algorithm, since all locust individuals in the population gather towards the optimal individual, the diversity of individuals within the population is lost, the algorithm converges prematurely, and a local optimal phenomenon occurs. In order to reduce the premature convergence of the genetic algorithm, the IGOA algorithm applies the logistic chaos map to the current optimal locust individual. The specific steps of the optimal search strategy are as follows:

[0157] The first step is to determine the variables Mapped to chaotic variables by the following formula :

[0158] ,

[0159] In the formula,

[0160] The second step is to substitute the above formula into the Logistic map for iteration to generate a chaotic variable sequence.

[0161] , is the maximum number of iterations.

[0162] Step 3: Chaos variables , which is converted into decision variables according to the following formula ,

[0163] ,

[0164] In the formula, represents the position of the kth locust at the tth iteration of the dth dimension,

[0165] Step 4: Evaluate decision variables Adaptability, update the current optimal individual,

[0166] The fifth step is to determine whether the maximum number of chaotic search is reached. If so, the chaotic search ends, otherwise, return to the second step.

[0167] Specifically, the SVDD model in S01 uses the hypersphere formed by the support vectors as the boundary line, the data within the boundary is called training samples, and the data outside the boundary is called other types of data. The decision function is used to determine whether the test data and the training data are of the same type. When the decision function is 1 and the distance between the test data and the center of the hypersphere is smaller than the radius of the hypersphere, it means that the test data and the training data are of the same type, otherwise they are of other types. The specific definition is as follows:

[0168] Assuming the radius of the hypersphere is R and the center is a, the expression of SVDD is:

[0169] ,

[0170] In the formula,

[0171] , is called the Lagrangian operator, is the support vector,

[0172] When classifying a given sample, calculate the distance between it and the hypersphere , and compared with the radius R of the hypersphere, the distance formula The details are as follows:

[0173] ,

[0174] The above formula can be used to determine whether it is a known working condition or an unknown working condition.

[0175] Specifically, the SVDD model uses G-means as the fitness function, and the specific expression is as follows:

[0176] ,

[0177] The calculation formula of G-mean is as follows:

[0178] ,

[0179] In the formula,

[0180] TP represents the number of correct classifications in the positive category.

[0181] FN represents the number of errors in the positive category.

[0182] TN represents the number of correct negative categories.

[0183] FP represents the number of incorrect classifications in the negative category.

[0184] Embodiment 3

[0185] Specifically, the steps of establishing the diagnosis model based on IGOA-SVDD in S03 are:

[0186] S031: Initialize algorithm parameters and set population size , maximum number of iterations ;

[0187] S032: Initialize the locust population algorithm using the reverse chaos learning algorithm;

[0188] S033: Use the inertia weight of the cloud model to divide the locust population into three categories: optimal population, suboptimal population and general population, optimize the penalty factor C and kernel parameter δ to be optimized, and calculate the fitness value of the locust individuals;

[0189] S034: Compare the current optimal result with the optimal result of the previous step optimization. If the current result is better than the previous step optimization result, the current fitness value is taken as the optimal value.

[0190] S035: Use formula (4) to calculate the probability of locust individuals to avoid falling into the local optimum;

[0191] S036: Determine whether the final condition is met. If the final condition is met, input the parameters to be optimized into the SVDD model. Otherwise, return to step S032.

[0192] Specifically, in S04, the PSO-VMD decomposition method is used to decompose the vibration signal when diagnosing and analyzing the fault, and the vibration signal is extracted from the time domain, frequency domain, energy and composite multi-scale weighted permutation entropy respectively to form a multi-dimensional feature fusion matrix. The vibration signal represented by N samples is collected for a total of S segments. The feature matrix can be expressed as:

[0193]

[0194] In the formula,

[0195] J=s+k+15 represents the dimension of the feature matrix.

[0196] S represents the number of locust algorithms,

[0197] Considering the length and complexity of data calculation, the mRMR principle is used to screen the vibration signal, and the reduced feature matrix is ​​defined as follows:

[0198]

[0199] Specifically, the optimal individual chaotic search strategy is used to find the optimal solution in S033.

[0200] Embodiment 4

[0201] In order to verify the effectiveness of this diagnostic method.

[0202] The diagnostic method is compared with the particle swarm optimization (PSO), genetic algorithm (GA), moth-flame optimization algorithm (MFO), grey wolf algorithm (GWO) and GOA. The Ackely function and Griewank function are used as benchmark functions, and the number of iterations is 500. The specific expression of the Ackely benchmark function is as follows:

[0203] ,

[0204] In the formula, x∈[-32,32], the dimension d=30, and the optimal solution is 0.

[0205] The specific expression of the Griewank benchmark function is as follows:

[0206] ,

[0207] In the formula, x∈[-600,600], dimension d=30, and the optimal solution is 0.

[0208] right and Thirty independent experiments were performed, as follows Figure 3 and Figure 4 As shown,

[0209] exist In the benchmark function, MFO has the worst convergence among all algorithms, and IGOA has the best convergence among the six algorithms, followed by GA, GWO, PSO, MFO, and GWO;

[0210] In the f2 test function, IGOA is still the best among the six algorithms, reaching the optimal value in 350 iterations, followed by GA, GWO, PSO, and MFO. GWO has the worst convergence among all algorithms.

[0211] pass and The performance test shows that the improved GOA algorithm is significantly better than other methods in terms of optimization ability, convergence speed and optimization accuracy.

[0212] Embodiment 5

[0213] In terms of fault identification, considering that it is relatively difficult to obtain fault samples in the early stage of high-voltage circuit breaker diagnosis, fault diagnosis is first performed from the normal state. A high-voltage circuit breaker fault diagnosis model is established using normal state data, and faults are identified as unknown states. 80% of the data under normal operating conditions are randomly selected as training samples, and the remaining 20% ​​are used as test samples. The remaining normal state and fault samples are used as test samples, and the calculation results are obtained after 30 operations. The fault diagnosis results of different methods are compared. Figure 5 shown.

[0214] IGOA-SVDD is compared with PSO-SVDD, GA-SVDD, MFO-SVDD, GWO-SVDD, GOA-SVDD and SVDD algorithms respectively. The diagnosis results are as follows Figure 5 shown.

[0215] It can be seen from the figure that IGOA-SVDD has the highest diagnostic accuracy among the diagnostic methods that establish diagnostic models based on normal operating conditions, and from the comparison results of G-means in the figure, it can be seen that IGOA-SVDD has the highest comparison results of G-means among all methods.

[0216] In terms of unknown fault diagnosis, loose contacts, detached contacts, and burned contacts are respectively regarded as unknown faults, and the fault categories are recorded as fault I, fault II, and fault III, respectively. These three fault experiments are recorded as condition a, condition b, and condition c, respectively. The same algorithm is used to identify unknown faults, and the training and testing strategy sample setting strategy for unknown faults is: similar to normal working conditions, 80% are randomly selected from the normal state, fault I, fault II, and fault III as training sets, and all the data in the training sets are marked as 1 as known faults. The remaining samples and unknown faults are used as test samples, and each algorithm is run an average of 30 times. The diagnosis results are as follows: Figure 6 shown.

[0217] from Figure 6 It can be seen that the diagnostic accuracy of the IGOA algorithm under the three conditions reached the highest value, which is higher than the other six algorithms.

[0218] In order to further analyze the effectiveness of the method proposed in this paper, the average value of unknown fault diagnosis under three conditions a, b, and c is calculated, as shown in Table 1.

[0219] Table 1 Average diagnostic results under three conditions

[0220]

[0221] It can be clearly concluded from Table 1 that IGOA-SVDD has the highest diagnostic accuracy among all methods.

[0222] In known-state fault identification, the average accuracy of IGOA-SVDD is 18.55%, 16.13%, 8.62%, 8.42%, 4.83% and 7.13% higher than that of SVDD, PSO-SVDD, GA-SVDD, MFO-SVDD, GWO-SVDD and GOA-SVDD, respectively.

[0223] In the identification of unknown fault states, the average identification accuracy of IGOA-SVDD is 16.22%, 12.03%, 5.4%, 6.22%, 7.32% and 8.56% higher than that of SVDD, PSO-SVDD, GA-SVDD, MFO-SVDD, GWO-SVDD and GOA-SVDD, respectively.

[0224] Meanwhile, IGOA-SVDD has the highest mean fault diagnosis accuracy (G-Means) among the three scenarios, which is 18.58%, 15.22%, 8.29%, 8.66%, 5.68% and 8.22% higher than SVDD, PSO-SVDD, GA-SVDD, MFO-SVDD, GWO-SVDD and GOA-SVDD, respectively.

[0225] In summary, IGOA-SVDD can effectively solve the problem of high-voltage circuit breaker position fault diagnosis, and its classification accuracy is significantly higher than that of the other five methods.

[0226] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0227] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A circuit breaker unknown fault diagnosis method based on IGOA-SVDD, characterized in that: The following steps are involved: S01: The unknown fault diagnosis problem is regarded as an abnormal data detection problem and the support vector data description (SVDD) algorithm is used for modeling; S02: Aiming at the problem that the locust algorithm (GOA) is prone to fall into the local optimum, the cloud model inertia weight and chaos reverse learning algorithm are used to improve the locust algorithm, and an improved locust algorithm (IGOA) is proposed; S03: Aiming at the problem that the penalty factor and kernel function in the support vector data description algorithm are difficult to determine, the improved locust algorithm is used to optimize the support vector data description algorithm, and a diagnostic model based on IGOA-SVDD is proposed; S04: Use the IGOA-SVDD diagnostic model to diagnose unknown faults of 110kV circuit breakers and obtain diagnostic results.

2. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 1, characterized in that: The steps of the improved locust algorithm in S02 are as follows: S021: Use the chaotic reverse learning strategy to initialize the population so that the population is distributed as evenly as possible in the feasible solution space, improving the uniformity and diversity of the initial population distribution of the locust algorithm; S022: Introduce the inertia weight cloud model into the locust algorithm and use different inertia weight strategies to adjust the convergence speed of the algorithm; S023: Based on the chaotic logistic mapping principle, local deep search is performed to optimize the later iterations of the locust algorithm, reduce the probability of falling into the local optimum, specify the optimal fitness value, and execute the first two steps again until the end standard is reached.

3. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 2, characterized in that: In S021, the population initialization based on the Logistic chaotic adversarial learning strategy is adopted to improve the uniformity and diversity of the initial population distribution of the locust algorithm in two stages, specifically: In the first stage, the chaotic sequence is generated by using the logistic mapping of formula (1) to initialize the population: Formula (1) In the formula, represents the chaotic variable, represents the sequence number of the chaotic variable, represents the sequence number of the population, r represents the dimension of the population; In the second stage, we use equations (2) and (3) to obtain a new population, calculate and sort the new population, and select the best N individuals as the initial population. The specific calculation process is: Simultaneously calculate and evaluate a candidate solution and the corresponding matching inverse solution, select the optimal solution as the next generation individual, and , calculate the opposite solution of the relevant match according to the following formula ,and ( represents the upper boundary, represents the lower boundary), calculated as follows: ， Convert the above formula from one dimension to high dimension as follows: ， Map the chaotic sequence into the solution space, Get population , Formula (2) Population individuals , Locust The first The initial value of the dimension, and To search for upper and lower bounds, By population Calculate the reverse population , The reverse population is expressed as: Formula (3) The population and Merge to get a new population.

4. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 2, characterized in that: In the S022, the inertia weight of the cloud model is used to divide the locust population into good sub-groups, ordinary sub-groups and bad sub-groups. Different sizes of inertia weights are used to improve the convergence speed and expand the search range. The modified mathematical model is expressed as follows: Formula (4) In the formula, , Respectively Locust The upper and lower bounds of the dimensional variables, is the target location of the locust swarm, is a linearly decreasing coefficient.

5. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 1, characterized in that: The SVDD model in S01 uses the hypersphere formed by the support vector as the boundary line. The data within the boundary is called training samples, and the data outside the boundary is called other types of data. The decision function is used to determine whether the test data and the training data are of the same type. When the decision function is 1 and the distance between the test data and the center of the hypersphere is smaller than the radius of the hypersphere, it means that the test data and the training data are of the same type, otherwise they are of other types. The specific definition is as follows: Assuming the radius of the hypersphere is R and the center is a, the expression of SVDD is: ， In the formula, , is called the Lagrangian operator, is the support vector, When classifying a given sample, calculate the distance between it and the hypersphere , and compared with the radius R of the hypersphere, the distance formula The details are as follows: ， The above formula can be used to determine whether it is a known working condition or an unknown working condition.

6. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 1, characterized in that: The SVDD model uses G-means as the fitness function, and the specific expression is as follows: ， The calculation formula of G-mean is as follows: ， In the formula, TP represents the number of correct classifications in the positive category. FN represents the number of errors in the positive category. TN represents the number of correct negative categories. FP represents the number of incorrect classifications in the negative category.

7. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 1, characterized in that: The steps for establishing the diagnostic model based on IGOA-SVDD in S03 are: S031: Initialize algorithm parameters and set population size , maximum number of iterations ; S032: Initialize the locust population algorithm using the reverse chaos learning algorithm; S033: Use the inertia weight of the cloud model to divide the locust population into three categories: optimal population, suboptimal population and general population, optimize the penalty factor C and kernel parameter δ to be optimized, and calculate the fitness value of the locust individuals; S034: Compare the current optimal result with the optimal result of the previous step optimization. If the current result is better than the previous step optimization result, the current fitness value is taken as the optimal value. S035: Use formula (4) to calculate the probability of locust individuals to avoid falling into the local optimum; S036: Determine whether the final condition is met. If the final condition is met, input the parameters to be optimized into the SVDD model. Otherwise, return to step S032.

8. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 1, characterized in that: When diagnosing and analyzing the fault in S04, the PSO-VMD decomposition method is used to decompose the vibration signal, and the vibration signal is extracted from the time domain, frequency domain, energy and composite multi-scale weighted permutation entropy respectively to form a multi-dimensional feature fusion matrix. The vibration signal represented by N samples is collected for a total of S segments. The feature matrix can be expressed as: ， In the formula, J=s+k+15 represents the dimension of the feature matrix. S represents the number of locust algorithms, Considering the length and complexity of data calculation, the mRMR principle is used to screen the vibration signal, and the reduced feature matrix is ​​defined as follows: 。 9. The circuit breaker unknown fault diagnosis method based on IGOA-SVDD according to claim 7, characterized in that: In S033, the optimal individual chaos search strategy is used for optimization.

Images