A method for locating a direct current insulation fault in a power plant

Abstract

The application discloses a method for locating insulation fault of direct current in a thermal power plant. Firstly, the branch with insulation fault is determined based on insulation resistance. Then, the collected signals are subjected to feature extraction through wavelet transform, and the support vector machine is combined to perform fault classification and identification. Then, a positioning solution algorithm based on particle swarm optimization is utilized, and the similarity of signal features is taken as a target function to realize high-precision fault location. Finally, the fault location result is verified. When the positioning result fails to pass the verification, the fault location process can be automatically re-executed, and a parameter optimization algorithm based on particle swarm optimization is combined with matching degree constraint as a target function to adjust and optimize the parameters of the positioning solution algorithm. The application can quickly and accurately locate the insulation fault point of the direct current system in the thermal power plant through multi-dimensional feature analysis algorithm, and improve the accuracy of fault classification.

Description

Technical Field

[0001] This invention relates to the field of thermal power generation technology, specifically to a method for locating DC insulation faults within a thermal power plant. Background Technology

[0002] In the power systems of thermal power plants, the stable operation of the DC system is crucial. However, it is susceptible to equipment aging and environmental changes during long-term operation, leading to frequent insulation faults. Traditional insulation fault detection methods (such as insulation monitoring circuits) can only issue alarm signals when the insulation resistance is lower than a set value, but they cannot accurately locate the fault point. Moreover, manual inspection is inefficient, as checking each line and piece of equipment is time-consuming and labor-intensive. Although fault location technologies using traveling wave signals and injection signals have emerged, these methods are easily interfered with in the complex electromagnetic environment of thermal power plants, making it difficult to guarantee the accuracy and reliability of the location. In addition, existing methods also have shortcomings in fault classification and identification, making it difficult to accurately distinguish between high-resistance grounding faults, low-resistance grounding faults, and intermittent grounding faults, resulting in low fault handling efficiency. Summary of the Invention

[0003] The present invention aims to solve the problems of poor positioning accuracy and reliability, as well as low fault handling efficiency of existing insulation fault detection methods, and provides a method for locating DC insulation faults in thermal power plants.

[0004] To solve the above problems, the present invention is achieved through the following technical solution:

[0005] A method for locating DC insulation faults within a thermal power plant includes the following steps:

[0006] Step 1: Perform real-time insulation monitoring on each branch of the DC system in the thermal power plant and obtain the insulation resistance of each branch. When the insulation resistance of a branch is detected to be lower than the preset threshold, it is determined that there is an insulation fault in that branch.

[0007] Step 2: Apply a pulse detection signal to the insulation fault branch, and simultaneously inject low-frequency auxiliary signals between the positive and negative buses of the DC system and ground.

[0008] Step 3: Simultaneously acquire pulse detection signals and low-frequency auxiliary signals at the detection points of the insulation fault branch;

[0009] Step 4: Use multi-scale wavelet transform to decompose the pulse detection signal and low-frequency auxiliary signal collected from the insulation fault branch, and extract the signal features at different scales.

[0010] Step 5: Integrate the signal features extracted from different scales by wavelet transform, construct a signal feature vector, and input the signal feature vector into a fault classification model based on support vector machine to classify and identify the fault type of the insulation fault branch.

[0011] Step 6: Based on the fault type of the insulation fault branch, combined with its electrical topology and line parameters, use the particle swarm optimization-based localization algorithm to solve the problem. Let each particle position in the particle swarm represent a possible fault location. Use the similarity of signal features as the objective function to iteratively optimize and solve the fault location of the insulation fault branch, and finally determine the precise location of the fault.

[0012] Step 7: After determining the location of the insulation fault, use a partial discharge detector to perform partial discharge detection near the insulation fault point and verify the location results.

[0013] If partial discharge is detected and the characteristics of the partial discharge match the fault type, then the accuracy of the fault location result is verified.

[0014] If no partial discharge phenomenon is detected or the partial discharge characteristics do not match, step 6 is repeated. In each iteration of the localization solution algorithm, a parameter optimization algorithm based on particle swarm optimization is used. Each particle position in the particle swarm represents a possible key parameter vector. The partial discharge characteristic matching degree constraint is used as the objective function to iteratively optimize the parameter vector of the localization solution algorithm. Finally, the optimized parameter vector is determined, and the parameters of the localization solution algorithm are adjusted and optimized.

[0015] In step 1 above, the insulation resistance of the branch is obtained by connecting insulation monitoring circuits in parallel between the positive terminal and ground, and between the negative terminal and ground, respectively. The insulation resistance R is:

[0016]

[0017] In the formula, U0 is the rated voltage of the DC system, and R b i is the internal resistance of the insulation monitoring circuit. + i is the current between the positive terminal of the branch and ground measured by the insulation monitoring circuit. - The current between the negative terminal of the branch and ground is measured by the insulation monitoring circuit.

[0018] In step 2 above, both the pulse detection signal and the low-frequency auxiliary signal are generated by a signal generator, wherein:

[0019] The pulse detection signal u applied to the insulation fault branch pulse for:

[0020]

[0021] The low-frequency auxiliary signal u1 injected between the positive bus and ground of the DC system is:

[0022]

[0023] The low-frequency auxiliary signal u2 injected between the negative bus and ground of the DC system is:

[0024]

[0025] In the formula, Up is the amplitude of the pulse detection signal, n is a natural number, T is the pulse period, τ is the pulse width; U1 is the amplitude of the low-frequency auxiliary signal, ω1 is the angular frequency of the low-frequency auxiliary signal, and t is time. and These are the initial phases of the injected signals for the positive and negative buses, respectively. This represents the phase difference.

[0026] In step 3 above, the detection point of the branch is located at the beginning of the branch, the intersection, or the device access point.

[0027] In step 4 above, the signal features include the signal energy, kurtosis, root mean square, and wavelet energy entropy.

[0028] In step 5 above, the fault type of the insulation fault branch includes high-resistance grounding fault, low-resistance grounding fault, or intermittent grounding fault.

[0029] In step 6 above, in the localization algorithm based on particle swarm optimization, the objective function of particle i is Loss(X). i )for:

[0030]

[0031] In the formula, Let X be the position vector of the particle. i The theoretical value of the signal characteristic f at point F actual w is the measured value of the signal characteristic f at the detection point of the current insulation fault branch. f Here, F represents the weighting coefficients of feature f, where F is the number of signal features, i = 1, 2, ..., N. p N p The particle swarm size is used to locate the solution algorithm.

[0032] In step 7 above, if the matching degree of the partial discharge feature is less than the matching degree threshold of the fault type, it is determined that the partial discharge feature and the fault type match or do not match; otherwise, it is determined that the partial discharge feature and the fault type match.

[0033] The matching degree M of the above partial discharge characteristics is:

[0034]

[0035] In the formula, f pdi f is the measured value of the characteristic parameter i of partial discharge detected by the partial discharge detector. thiis the theoretical value of the characteristic parameter i of the partial discharge corresponding to the fault type, and n is the number of characteristic parameters of the partial discharge.

[0036] In step 7 above, in the parameter optimization algorithm based on particle swarm optimization, the objective function of particle l is Loss′(X). l )for:

[0037]

[0038] In the formula, d k The fault location d is calculated by the k-th iteration of the currently running localization algorithm. k,th The precise location of the fault point calculated by the previous fault location algorithm is given by: K, the maximum number of iterations of the fault location algorithm, M, the partial discharge feature matching degree, and λ, where l = 1, 2, ..., N. q N q The particle swarm size is used to optimize the parameters of the algorithm.

[0039] Compared with the prior art, the present invention has the following characteristics:

[0040] 1. Fast and accurate location: By combining the similarity of signal features with a location solution algorithm based on particle swarm optimization, the fault point can be located quickly and accurately. Compared with traditional methods, the fault diagnosis time is greatly shortened, and the power outage time and economic losses caused by the fault are reduced.

[0041] 2. Efficient Fault Classification: Employing wavelet transform-based feature extraction and support vector machine-based fault classification modules, it can accurately distinguish between high-resistance grounding faults, low-resistance grounding faults, and intermittent grounding faults, thereby improving the efficiency and targeting of fault handling.

[0042] 3. Strong anti-interference capability: By decomposing the signal through wavelet transform and extracting signal features at multiple scales, the anti-interference capability in complex electromagnetic environments is effectively improved, ensuring the reliability of fault location.

[0043] 4. Reliable verification process: An additional step was added to verify the location results. Partial discharge detection was used to further confirm the location of the fault point, which enhanced the accuracy and reliability of the entire fault location method and avoided misjudgment.

[0044] 5. Adaptive optimization: When the location result verification fails, the fault location process can be automatically re-executed. The parameter optimization algorithm based on particle swarm optimization is combined with the matching degree constraint as the objective function to adjust and optimize the parameters of the location solution algorithm, thereby improving the system's adaptability and fault tolerance. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific examples.

[0046] A method for locating DC insulation faults within a thermal power plant, comprising the following steps:

[0047] S1: Real-time insulation monitoring of each branch of the DC system within the thermal power plant is performed to obtain the insulation resistance of each branch. When the insulation resistance of a branch is detected to be lower than the preset threshold, it is determined that there is an insulation fault in that branch.

[0048] S11: Insulation monitoring circuits are connected in parallel between the positive terminal and ground, and between the negative terminal and ground, for each branch of the DC system within the thermal power plant.

[0049] The insulation monitoring circuit described above is the same insulation monitoring circuit used in traditional insulation fault detection in thermal power plants, which uses the DC superposition method for insulation monitoring.

[0050] S12: Inject weak AC signals of the same amplitude into the positive and negative buses of the DC system respectively, and obtain the voltage u of the positive and negative buses of the DC system after superimposing the AC signals. + and u _ Its expression is:

[0051]

[0052] In the formula, U0 is the rated voltage of the DC system, U is the amplitude of the superimposed AC signal, and ω is the angular frequency of the AC signal. Let t be the initial phase of the AC signal, and t be time.

[0053] S13: Measure the current i between the positive terminal and ground, and between the negative terminal and ground, using an insulation monitoring circuit. + and i - Its expression is:

[0054]

[0055] In the formula, R + and R - These are the insulation resistances between the positive terminal and ground, and between the negative terminal and ground, respectively. b This is the internal resistance of the insulation monitoring circuit.

[0056] S14: Based on current i + and i - Calculate the insulation resistance R of each branch using the following formula:

[0057]

[0058] S15: Define the preset threshold as Rth (Unit is kΩ), when the calculated R <R th If the condition is met, it is determined that there is an insulation fault in the branch; otherwise, there is no insulation fault in the branch.

[0059] S2: Apply a pulse detection signal of a specific frequency to the branch with insulation fault (insulation fault branch), and simultaneously inject low-frequency auxiliary signals with a specific phase relationship between the positive and negative buses of the DC system and ground.

[0060] After identifying a branch with an insulation fault, a pulse detection signal u is applied to that branch. pulse Its expression is:

[0061]

[0062] In the formula, Up is the amplitude of the pulse detection signal (in V), t is time, n is a set natural number, T is the pulse period (in s), and τ is the pulse width (in s).

[0063] Simultaneously, a low-frequency auxiliary signal u1 (in Hz) is injected between the positive bus and ground of the DC system, and a low-frequency auxiliary signal u2 (in Hz) is injected between the negative bus and ground. The expression for this is:

[0064]

[0065] In the formula, U1 is the amplitude of the low-frequency auxiliary signal (in V), and ω1 is the angular frequency of the low-frequency auxiliary signal. and These are the initial phases of the injected signals for the positive and negative buses, respectively, and they satisfy a specific phase relationship, namely: in The phase difference is set in degrees, and t is time.

[0066] Both the pulse detection signal and the low-frequency auxiliary signal are generated by a high-precision signal generator. When injecting the pulse detection signal and the low-frequency auxiliary signal, an isolation transformer is used for electrical isolation to ensure the safety of the signal injection. The turns ratio of the isolation transformer is n1:n2, where n1 is the number of turns on the primary side and n2 is the number of turns on the secondary side.

[0067] S3: Simultaneously acquire pulse detection signals and low-frequency auxiliary signals at the detection point of the insulation fault branch.

[0068] Detection points for branch circuits are typically located at critical nodes, such as the branch's starting point, intersections, or equipment connection points. The acquisition of pulse detection signals and low-frequency auxiliary signals is achieved using traditional electrical sensors, such as voltage transformers (PTs), current transformers (CTs), and phase detectors. The transformation ratios of PTs and CTs are np and ni, respectively, and the accuracy of the phase detector is... (Unit: degrees) to improve the accuracy and coverage of signal acquisition.

[0069] S4: Utilizing multi-scale wavelet transform to analyze the pulse detection signal x acquired from the insulation fault branch. pulse and low-frequency auxiliary signal x low The signal is decomposed and its features at different scales are extracted.

[0070] Wavelet transform is a signal processing technique that decomposes a signal into frequency components at different scales, enabling multi-scale signal analysis. The basic principle of wavelet transform is to perform an inner product operation between the signal and a wavelet function to obtain the signal's frequency components at different scales and locations. The multi-scale wavelet transform coefficients are also known as approximation coefficients cA. J and details coefficient cD j They are represented as follows:

[0071]

[0072] cD j = <x,ψ j,k >

[0073] In the formula, x is the signal to be decomposed. and ψ j,k These are the translational and scaling versions of the scaling function and wavelet function at position k in scale J, respectively; j = 1, 2, ..., J, where J is the number of decomposition scales, typically ranging from [3, 5]; <*, *> represent the inner product operation, used to indicate the correlation between the signal and the wavelet basis function.

[0074] Signal features, including signal energy E, are extracted based on multi-scale wavelet transform. j , kurtosis K j Root Mean Square (RMS) j and wavelet energy entropy S we .

[0075] The energy of a signal measures its strength. At each scale j, the signal energy E... j for:

[0076]

[0077] The kurtosis of a signal measures the degree of abnormal peaks in the signal. At each scale j, the kurtosis K of the signal... j for:

[0078]

[0079] The root mean square (RMS) of a signal measures the statistical intensity of signal or data fluctuations. At each scale j, the RMS of the signal is... j for:

[0080]

[0081] The wavelet energy entropy of a signal measures the complexity of its energy distribution. At all scales, the wavelet energy entropy S of a signal... we for:

[0082]

[0083] In the above formula, cD j [k] is the detail coefficient of scale j at position k, j = 1, 2, ..., J, where J is the number of decomposition scales, and k = 1, 2, ..., K, where K is the number of positions.

[0084] S5: Integrate the signal features extracted from different scales by wavelet transform to construct the signal feature vector F. svm and the signal feature vector F svm Input a fault classification model based on support vector machine (SVM) to classify and identify the fault types of insulation fault branches.

[0085] S51: The signal feature vector F constructed by integrating signal features at different scales. svm for:

[0086] F svm =[E1,E2,...,E J ;K1,K2,...,K J ;RMS1,RMS2,...,RMS J S we ]

[0087] In the formula, E j Let K be the signal energy at scale j. j Let RMS be the signal kurtosis at scale j. j S is the root mean square of the signal at scale j. we Let J be the wavelet energy entropy, and J be the number of wavelet decompositions.

[0088] S52: Pre-training of a fault classification model based on Support Vector Machine (SVM).

[0089] Collect a large amount of sample data on known fault types, including signal feature vectors of high-resistance grounding faults, low-resistance grounding faults, and intermittent grounding faults. And label the corresponding fault type y (m) ∈{1,2,3}. Where m=1,2,...,M, M is the number of samples, label 1 represents high-resistance grounding fault, label 2 represents low-resistance grounding fault, and label 3 represents intermittent grounding fault.

[0090] Radial basis functions (RBFs) are chosen as the kernel function of SVM, and the optimization problem of SVM is constructed by setting constraints. Wherein:

[0091] The kernel function of SVM is:

[0092]

[0093] The constraints are:

[0094]

[0095] The optimization problem for constructing an SVM is:

[0096]

[0097] In the above formula, σ is the kernel function width parameter, with a value range of [0.1, 10], w is the weight vector; b is the bias term; C is the penalty parameter, with a value range of [1, 1000]. For kernel function mapping, ξ (m) It is a slack variable.

[0098] The Sequence Minimum Optimization (SMO) algorithm is used to solve the above optimization problem, and the parameters (weight vector w) of the fault classification model are obtained. opt and bias term b opt This completes the pre-training of the fault classification model, resulting in a fault classification model capable of fault classification and identification.

[0099] S53: Transfer the signal feature vector F of the insulation fault branch svm The input is fed into the trained fault classification model, which then calculates the decision function f(F). svm The fault type is determined based on the magnitude and sign of the decision function value.

[0100] The decision function f(F) of the fault classification model svm )for:

[0101]

[0102] If f(F) svm If )>0, then the fault classification model determines that the insulation fault branch is a high-resistance grounding fault;

[0103] If f(F) svmIf ) < 0, then the fault classification model determines that the insulation fault branch is a low-resistance grounding fault;

[0104] If f(F) svm If ) = 0, then the fault classification model determines that the insulation fault branch is an intermittent grounding fault.

[0105] S6: Based on the fault type of the insulation fault branch, combined with its electrical topology and line parameters, a particle swarm optimization-based localization algorithm is used to iteratively optimize and solve the fault point location of the insulation fault branch with the similarity of signal features as the objective function, and finally determine the precise location of the fault point.

[0106] S61: Initialization Phase

[0107] ① Create a particle swarm with a size of N p A particle swarm, with the position X of each particle randomly initialized. i and speed V i Thus, the initial particle swarm is obtained.

[0108] The position X of each particle in the particle swarm i Represents a possible location of a fault, i = 1, 2, ..., N p During random initialization, the particle's position X i The value range is [0,1] times the length L of the current insulation fault branch, and the particle velocity V i The range of values ​​is [-V max V max ], where V max To achieve the maximum speed, it is typically set to 0.1 times the length of the current insulation fault branch per iteration step.

[0109] ② Define the objective function:

[0110]

[0111] In the formula, Let X be the position vector of the particle. i The theoretical value of the signal characteristic f at point F; actual The measured value of the signal characteristic f at the detection point of the current insulation fault branch is obtained by analyzing the signal (pulse detection signal x) collected on-site. pulse and low-frequency auxiliary signal x low The signal characteristics at the time of the fault were obtained through feature extraction; f Here, F represents the weighting coefficients of feature f, where F is the number of signal features, i = 1, 2, ..., N. p N pThe particle swarm size is used to locate the solution algorithm. In this embodiment, the signal feature f includes the signal energy, the signal kurtosis, the signal root mean square, and the signal wavelet energy entropy.

[0112] ③ Take the position of a particle in the initial particle swarm as the initial individual optimal position P of that particle. best,i And its corresponding objective function value is used as the initial individual optimal objective function value Loss(P) best,i ).

[0113] ④ Find the position of the particle corresponding to the optimal objective function value of the initial particle swarm as the initial global optimal position G. best And its corresponding objective function value is used as the initial global optimal objective function value Loss(G). best ).

[0114] S62: Particle Swarm Optimization Iterative Stage

[0115] ① Update the position and velocity of each particle in the current particle swarm:

[0116] The velocity update formula for particle i is:

[0117]

[0118] The formula for updating the position of particle i is:

[0119]

[0120] In the formula, Let i be the velocity of particle i after the update. The velocity of particle i before the update; The position of particle i after the update. The position of particle i before updating; w is the inertia weight, which controls the degree to which the particle depends on the previous velocity, usually decreasing linearly from 0.9 to 0.4; c1 and c2 are cognitive and social learning factors, usually taking c1 = c2 = 2; r1 and r2 are random numbers in [0,1]. Let be the optimal position of particle i. This is the globally optimal position.

[0121] ② Boundary handling: If the position of particle i is updated Beyond the physical range of the branch (e.g.) or If the position is set to a boundary value, it will be restricted to [0,L] and a certain penalty will be imposed (such as reducing its objective function value).

[0122] ③ Individual optimal update: Calculate the position of particle i after the update. objective function value like The value of the objective function is greater than the current individual's optimal objective function value, Loss(P). best,i If the current individual's optimal position P is determined, then... best,i Update to the position of particle i after the update. The current optimal individual objective function value Loss(P) best,i Updated to Otherwise, the current optimal position P of the individual best,i and the current individual's optimal objective function value Loss(P) best,i )constant.

[0123] ④ Global Update: After completing the optimal update for all individuals, find the optimal objective function value of the current particle swarm and compare it with the current global objective function value Loss(G). best Compare: If the optimal objective function value of the current particle swarm is greater than the current global objective function value, Loss(G) best If the current global optimal position G is determined, then the global optimal position G will be determined. best Update the position of the particle corresponding to the optimal objective function value of the current particle swarm, and set the current global optimal objective function value Loss(G) to the position of the particle. best Update the current global optimum position G to the optimal objective function value of the current particle swarm; otherwise, update the current global optimum position G. best and the current globally optimal objective function value Loss(G) best )constant.

[0124] ⑤ Set the maximum number of iterations T max (e.g., 100 times) or the convergence threshold f of the objective function th (e.g., 0.99). When the number of iterations reaches T... max Or the maximum objective function value f(G) best )≥f th Stop iteration when the time is reached; output the final global optimal position G. best The precise location of the insulation fault point is used; otherwise, the particle swarm optimization iteration process of S62 is re-executed.

[0125] S7: After determining the location of the insulation fault point, a partial discharge detector is used to detect partial discharge near the insulation fault point to verify the location result: if a partial discharge phenomenon is detected and the partial discharge characteristics match the fault type, the accuracy of the fault location result is verified; if no partial discharge phenomenon is detected or the partial discharge characteristics do not match, the fault location process in S4 is re-executed, and the parameters of the location solution algorithm are adjusted and optimized in each iteration of the location solution algorithm.

[0126] S71: Calculate the partial discharge characteristic matching degree M:

[0127] Let F be the partial discharge feature vector detected by the partial discharge detector.pd for:

[0128] F pd =[f pd1 ,f pd2 ,...,f pd ]

[0129] Based on the pre-established mapping model between fault types and partial discharge characteristic vectors, the theoretical partial discharge characteristic vector F is calculated. th for:

[0130] F th =[f th1 ,f th 2,...,f thn ]

[0131] The actual detected partial discharge feature vector F pd Compared with the theoretical partial discharge eigenvector F th The matching degree M is:

[0132]

[0133] In the above formula, f pd1 ,f pd2 ,…,f pdn These are the actual values ​​of various characteristic parameters of partial discharge, f. th1 ,f th2 ,...,f thn These are the theoretical values ​​of various characteristic parameters of partial discharge. n is the number of characteristic parameters of partial discharge. The value of M ranges from [0,1]. In this embodiment, the characteristic parameters include pulse amplitude, pulse rise time, pulse half-peak time, pulse repetition rate, etc.

[0134] S72: Match the partial discharge characteristic matching degree M with the corresponding fault type matching degree threshold M. th Comparison: When M <M th If the partial discharge characteristics do not match the fault type, proceed to step S73; otherwise, determine if the partial discharge characteristics match the fault type.

[0135] S73: Using a parameter optimization algorithm based on particle swarm optimization, with the partial discharge feature matching degree constraint as the objective function, the parameter vector of the localization solution algorithm is iteratively optimized and solved, and finally the optimized parameter vector is determined.

[0136] S731: Initialization Phase

[0137] ① Create a particle swarm with a size of N q A particle swarm, with the position X of each particle randomly initialized. l and speed Vl Thus, the initial particle swarm is obtained.

[0138] The position X of each particle in the particle swarm l Represents a possible key parameter vector, l = 1, 2, ..., N q X l =[x1,x2,...,x e ], V l =[v1,v2,...,v e ].

[0139] x1,x2,...,x e These represent key parameters in the localization algorithm, such as weight coefficients, iteration step size, and convergence threshold; v1, v2, ..., v e The rate of change of each key parameter in the solution algorithm is located, such as the rate of change of the weight coefficient, the rate of change of the iteration step size, and the rate of change of the convergence threshold; e is the number of key parameters in the solution algorithm.

[0140] ② Define the objective function:

[0141]

[0142] In the formula, d k The fault location d is calculated by the k-th iteration of the currently running localization algorithm. k,th K represents the precise location of the fault point calculated by the previous location solving algorithm, M represents the maximum number of iterations of the current location solving algorithm, and λ represents the partial discharge feature matching degree.

[0143] ③ Take the position of a particle in the initial particle swarm as the initial individual optimal position P of that particle. best,l And its corresponding objective function value is used as the initial individual optimal objective function value Loss′(P) best,l ).

[0144] ④ Find the position of the particle corresponding to the optimal objective function value of the initial particle swarm as the initial global optimal position G. best And its corresponding objective function value is used as the initial global optimal objective function value Loss′(G). best ).

[0145] S732: Particle Swarm Optimization Iterative Stage

[0146] ① Update the position and velocity of each particle in the current particle swarm:

[0147] The velocity update formula for particle l is:

[0148]

[0149] The position update formula for particle l is:

[0150]

[0151] In the formula, Let l be the velocity of the updated particle. The velocity of particle l before the update; The position of particle l after the update. The position of particle l before updating; w is the inertia weight, which controls the degree to which the particle depends on the previous velocity, usually decreasing linearly from 0.9 to 0.4; c1 and c2 are cognitive and social learning factors, usually taking c1 = c2 = 2; r1 and r2 are random numbers in [0,1]. Let be the optimal position for particle l. This is the globally optimal position.

[0152] ② Individual optimal update: Calculate the position of particle l after the update. objective function value like Greater than the current individual's optimal objective function value Loss′(P) best,l If the current individual's optimal position P is determined, then... best,i Updated to the position of particle l after the update. The current optimal individual objective function value Loss′(P) best,l Updated to Otherwise, the current optimal position P of the individual best,l and the current individual's optimal objective function value Loss′(P) best,l )constant.

[0153] ③ Global Update: After completing the optimal update for all individuals, find the optimal objective function value of the current particle swarm and compare it with the current global objective function value Loss′(G best Compare: If the optimal objective function value of the current particle swarm is greater than the current global objective function value Loss′(G) best If the current global optimal position G is determined, then the global optimal position G will be determined. best Update the position of the particle corresponding to the optimal objective function value of the current particle swarm, and set the current global optimal objective function value Loss′(G) to the position of the particle. best Update the current global optimum position G to the optimal objective function value of the current particle swarm; otherwise, update the current global optimum position G. best and the current globally optimal objective function value Loss′(G best )constant.

[0154] ④ Set the maximum number of iterations N iter When the number of iterations reaches N iterStop iteration when the time is reached; output the final global optimal position G. best As the optimized parameter vector P opt Otherwise, the particle swarm optimization iteration process of S732 will be re-executed.

[0155] S733: Using the optimized parameter vector P opt Update the parameters in the localization algorithm and re-execute the algorithm until the partial discharge feature matching degree M ≥ M th .

[0156] In summary, this invention employs a multi-dimensional feature analysis algorithm to quickly and accurately locate fault points. First, it identifies branches with insulation faults based on insulation resistance. Then, it extracts features from the acquired signals using wavelet transform, and performs fault classification and identification using a support vector machine. Next, it utilizes a particle swarm optimization-based fault location algorithm, with the similarity of signal features as the objective function, to achieve high-precision fault location. Finally, the fault location results are verified. If the verification fails, the fault location process is automatically re-executed, and the parameters of the fault location algorithm are adjusted and optimized using a particle swarm optimization-based parameter optimization algorithm combined with matching degree constraints as the objective function. This invention can quickly and accurately locate insulation fault points in the DC system of a thermal power plant, improve the accuracy of fault classification, and ensure the reliability of the location results through verification and optimization processes. This effectively reduces power outage time and economic losses, and improves the operational stability of the DC system in thermal power plants.

[0157] It should be noted that although the embodiments described above are illustrative, they are not intended to limit the invention. Therefore, the invention is not limited to the specific embodiments described above. Any other embodiments obtained by those skilled in the art under the guidance of this invention without departing from its principles are considered to be within the protection scope of this invention.

Claims

1. A method for locating DC insulation faults within a thermal power plant, characterized in that, The steps include the following: Step 1: Perform real-time insulation monitoring on each branch of the DC system in the thermal power plant and obtain the insulation resistance of each branch. When the insulation resistance of a branch is detected to be lower than the preset threshold, it is determined that there is an insulation fault in that branch. Step 2: Apply a pulse detection signal to the insulation fault branch, and simultaneously inject low-frequency auxiliary signals between the positive and negative buses of the DC system and ground. Step 3: Simultaneously acquire pulse detection signals and low-frequency auxiliary signals at the detection points of the insulation fault branch; Step 4: Use multi-scale wavelet transform to decompose the pulse detection signal and low-frequency auxiliary signal collected from the insulation fault branch, and extract the signal features at different scales. Step 5: Integrate the signal features extracted from different scales by wavelet transform, construct a signal feature vector, and input the signal feature vector into a fault classification model based on support vector machine to classify and identify the fault type of the insulation fault branch. Step 6: Based on the fault type of the insulation fault branch, combined with its electrical topology and line parameters, use the particle swarm optimization-based localization algorithm to solve the problem. Let each particle position in the particle swarm represent a possible fault location. Use the similarity of signal features as the objective function to iteratively optimize and solve the fault location of the insulation fault branch, and finally determine the precise location of the fault. Step 7: After determining the location of the insulation fault, use a partial discharge detector to perform partial discharge detection near the insulation fault point and verify the location results. If partial discharge is detected and the characteristics of the partial discharge match the fault type, then the accuracy of the fault location result is verified. If no partial discharge phenomenon is detected or the partial discharge characteristics do not match, step 6 is repeated. In each iteration of the localization solution algorithm, a parameter optimization algorithm based on particle swarm optimization is used. Each particle position in the particle swarm represents a possible key parameter vector. The partial discharge characteristic matching degree constraint is used as the objective function to iteratively optimize the parameter vector of the localization solution algorithm. Finally, the optimized parameter vector is determined, and the parameters of the localization solution algorithm are adjusted and optimized.

2. The method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 1, the insulation resistance of the branch is obtained by connecting insulation monitoring circuits in parallel between the positive terminal and ground, and between the negative terminal and ground, respectively. for: In the formula, The rated voltage of the DC system. The internal resistance of the insulation monitoring circuit, The current between the positive terminal of the branch and ground is measured by the insulation monitoring circuit. The current between the negative terminal of the branch and ground is measured by the insulation monitoring circuit.

3. The method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 2, both the pulse detection signal and the low-frequency auxiliary signal are generated by a signal generator, wherein: Pulse detection signal applied to the insulation fault branch for: Low-frequency auxiliary signal injected between the positive bus and ground of the DC system for: Low-frequency auxiliary signal injected between the negative bus and ground of the DC system for: In the formula, The amplitude of the pulse detection signal. For natural numbers, For pulse period, The pulse width; The amplitude of the low-frequency auxiliary signal. The angular frequency of the low-frequency auxiliary signal. For time, and These are the initial phases of the injected signals for the positive and negative buses, respectively. , This represents the phase difference.

4. The method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 3, the detection points for the branch are located at the beginning of the branch, the intersection, or the device access point.

5. The method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 4, the signal features include the signal energy, kurtosis, root mean square, and wavelet energy entropy.

6. The method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 5, the fault types of the insulation fault branch include high-resistance grounding fault, low-resistance grounding fault, and intermittent grounding fault.

7. The method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 6, in the localization solution algorithm based on particle swarm optimization, the particles... objective function for: In the formula, The position vector of the particle Signal characteristics at the location The theoretical value, Signal characteristics at the detection point of the current insulation fault branch. The measured value, Features The weighting coefficients, The number of signal features, , The particle swarm size is used to locate the solution algorithm.

8. The method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 7, if the matching degree of the partial discharge feature is less than the matching degree threshold of the fault type, it is determined that the partial discharge feature and the fault type match or do not match; otherwise, it is determined that the partial discharge feature and the fault type match. The above partial discharge characteristic matching degree for: In the formula, Characteristic parameters of partial discharge detected by the partial discharge detector The measured value, Characteristic parameters of partial discharge corresponding to the fault type The theoretical value, This represents the number of characteristic parameters of partial discharge.

9. A method for locating DC insulation faults within a thermal power plant according to claim 1, characterized in that, In step 7, in the parameter optimization algorithm based on particle swarm optimization, particles... objective function for: In the formula, For the currently running localization algorithm, the th The location of the fault point calculated in the next iteration. The precise location of the fault point calculated by the previous location solving algorithm. To determine the maximum number of iterations for the localization algorithm, For partial discharge characteristic matching degree, For balance coefficient, , The particle swarm size is used to optimize the parameters of the algorithm.

Images