A power distribution network fault location method based on improved central collision optimization algorithm

Abstract

The application discloses a power distribution network fault location method based on an improved central collision optimization algorithm and belongs to the field of power distribution network fault diagnosis and intelligent optimization algorithm. The method models the fault location problem of a power distribution network containing a DG as a binary combination optimization problem, and makes three improvements to the standard binary central collision optimization algorithm. Firstly, a Hammersley low-dispersion sequence is used to replace the pseudo-random initialization to enhance the space coverage ability of the initial population. Secondly, an adaptive weighted covariance matrix estimation based on fitness ranking is introduced to improve the search efficiency in the decorrelation space. Thirdly, an elite-guided Cauchy segment exchange mutation strategy is designed to accelerate convergence while maintaining population diversity. Experimental results show that the method has a strong robustness and practicability, and the average location accuracy reaches 99.86% in 14 kinds of fault scenarios of the power distribution network containing a DG, including single-point fault, multi-point fault and information distortion fault, which is superior to other comparative algorithms.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network fault diagnosis and intelligent optimization algorithm technology, specifically involving a method for locating fault sections in power distribution networks based on an improved center collision optimization algorithm. Background Technology

[0002] With the large-scale integration of distributed generation into the distribution network, the traditional distribution network topology and power flow direction have undergone fundamental changes. The integration of distributed generation has transformed the distribution network from a single-source radial network into a complex multi-source network. The direction and magnitude of fault currents have become more complex, and the fault information reported by feeder terminal units (FTUs) may be distorted or even erroneous, posing a severe challenge to the location of fault sections in the distribution network.

[0003] Existing fault location methods for distribution networks mainly fall into two categories: analytical methods based on matrix analysis and methods based on intelligent optimization algorithms. Analytical methods achieve good location accuracy under ideal conditions with complete information, but their robustness is insufficient in complex scenarios such as FTU information distortion and changes in DG switching status. Methods based on intelligent optimization algorithms transform the fault location problem into a combinatorial optimization problem. By constructing a fitness function, they utilize metaheuristic algorithms to search for the optimal combination of fault sections. Commonly used algorithms include the Binary Particle Swarm Optimization (BPSO), Binary Grey Wolf (BGWO), and Binary Whale Optimization (BWOA). However, these algorithms suffer from slow convergence speed, susceptibility to local optima, and unstable location accuracy when facing high-dimensional binary search spaces, especially performing poorly in scenarios with multiple faults and information distortion.

[0004] The Centered Collision Optimizer (CCO) is a novel metaheuristic algorithm inspired by the head-on collision equation in classical physics. This algorithm has strong global search capabilities and the ability to escape local optima. However, when CCO is applied to the binary optimization problem of fault location in distribution networks, the following shortcomings still exist: (1) The population diversity of the standard random initialization is insufficient, which affects the global search capability of the algorithm; (2) The standard covariance estimation in the decorrelation space does not make full use of the information of excellent individuals in the population to guide the search direction; (3) The lack of a refined local search mechanism for the binary solution space leads to insufficient accurate location capability of the algorithm in the later stage.

[0005] Therefore, it is necessary to make targeted improvements to the CCO algorithm to enhance its accuracy and robustness in fault location problems in distribution networks with distributed generation (DG). Summary of the Invention

[0006] The method of the present invention includes the following steps: Step S1: Modeling the fault location problem in a distribution network with DG. Based on the distribution network topology, establish the upstream and downstream correlation matrices of each line segment and FTU switch, as well as the DG correlation matrix; model the fault location problem as a binary combinatorial optimization problem, using the fault state vector of the line segment... As decision variables, among them , Indicates the first One line malfunctioned. This indicates normal operation. The expected FTU state information is generated based on the fault state vector, and compared with the actually reported FTU information to construct a fitness function: in For the first Each FTU actually reports its status. The expected state calculated based on the failure assumptions. This represents a line fault state. The algorithm aims to minimize... To optimize the objective; when hour The theoretically optimal value indicates that the fault assumptions are completely consistent with the actual reported information and there are no redundant fault assumptions.

[0007] Step S2: Population initialization based on Hammersley low-difference sequences. The pseudo-random initialization in the original BCCO is replaced with Hammersley low-difference sequences. For a population size of... Dimensions The optimization problem is to generate the Hammersley quasi-random sequence matrix. Scaling it to the search space using a linear mapping : in , These are the lower and upper bound vectors for the decision variables, respectively. This is an element-wise multiplication. The first dimension of the Hammersley sequence is sampled using a uniform distribution. The remaining dimensions use Van der Corput sequences based on different prime bases, for the th... Dimension selection prime number , to integer Expand as Number base and fractional reversal: in for exist The digits in the number system. Hammersley sequences have low dissimilarity, which makes the initial population more evenly distributed in the search space, effectively avoiding search blind spots caused by the initial clustering of individuals and improving the algorithm's global exploration capability.

[0008] Step S3: Sigmoid binary conversion and initial fitness calculation. For each individual with a continuous value in the population, the Sigmoid transfer function is used to map it to... Interval: in This is the slope parameter of the Sigmoid function. Then, a threshold is used. Binarization yields the binary solution vector. :when hour ,otherwise Calculate the fitness value for each member of the binary population and record the global optimal solution. and its fitness .

[0009] Step S4: Fitness Sort and Spatial Grouping. Sort the population in ascending order according to fitness values ​​and initialize spatial allocation coefficients. Based on this, the population was divided into two groups: the de-related space renewal group and the original space renewal group, with the number of individuals in each group being respectively... and The two groups perform position updates in different spaces to balance the search resources of the algorithm in the decorrelation space and the original space.

[0010] Step S5: Adaptive weighted covariance matrix estimation and spatial transformation. Construct a ranking weight vector based on the current population's fitness ranking. , No. The weights of each individual (arranged in ascending order of fitness) are: The higher the ranking of an individual, the greater its weight. This is represented by a ranking-weight vector. Calculate the weighted mean and weighted covariance matrix: For the weighted covariance matrix Perform eigenvalue decomposition Obtain the transformation matrix Transform the population to a derelated space Compared to the standard covariance estimation in the original BCCO that assigns equal weights to all individuals, this method utilizes the ranking weighted information of the entire population, allowing superior individuals to contribute more to the covariance matrix, thereby guiding the search direction more accurately toward the global optimum.

[0011] Step S6: Center Collision Strategy Position Update. The center collision strategy is executed for both groups of individuals in different spaces. In the relevant space, the position update formula is: After the update, it is mapped back to the original space via inverse transformation. In the original space, the position update formula is: in Indicates from 1 to The index is randomly selected from the current individual to select individuals with a fitness better than the current individual; Indicates from 1 to A randomly selected index; For a binary mask vector, when hour ,otherwise ; Adaptive changes with iteration, This represents the current iteration number. This represents the maximum number of iterations. ; for Uniformly distributed random numbers within an interval.

[0012] Step S7: Binarization and Adaptive Update of Spatial Allocation Coefficients. Boundary correction is performed on the continuous solutions after position updates: if... or Then let The binary form is then generated using the Sigmoid function, and fitness is calculated. A greedy selection strategy is employed to retain the better individuals. The number of successful offspring generated from the relevant space group and the original space group is counted separately. and The space allocation coefficients are adaptively updated according to the following formula: And Limited to Within the range, that is, if Then let ,like Then let This ensures that both spaces always have sufficient search resources. (Based on the updated...) Re-determine the number of individuals to be allocated to the relevant space in the next round. .

[0013] Step S8: Elite-guided Cauchy segment exchange mutation. For each sorted individual... Perform a composite mutation combining elite guidance and Cauchy perturbation. Calculate the elite fragment length, which decreases with iteration: Randomly select the starting position of the segment and end position The global optimal solution exist Gene fragments within the specified range cover the corresponding locations of the current individual. Dimensions not covered by elite fragments... Calculate the Cauchy mutation probability that decreases with iteration: And generate Cauchy distributed random numbers: like Then perform a bit flip on that dimension. The multiplier of 10 is used to adjust the effective flip probability of the Cauchy distribution to be equal to... Magnitude matching is used to fully utilize the heavy-tailed characteristic of the Cauchy distribution to achieve long-distance jumps. The fitness of individuals after mutation is calculated. ,like If the mutation result is accepted, the population and the global optimum are updated; otherwise, the original individuals remain unchanged. This mutation strategy rapidly propagates superior gene patterns in the early stages through larger elite fragment lengths, and maintains population diversity in the later stages through smaller fragment lengths and Cauchy perturbations, balancing convergence speed and global search capability.

[0014] Step S9: Iteration Termination Check and Result Output. Determine if the maximum number of iterations has been reached. If the target is not reached, return to step S4 and continue iterating; if the target is reached, output the globally optimal binary solution vector. The line segment corresponding to the position with a value of 1 is the fault location result.

[0015] Compared with the prior art, the present invention has the following advantages: (1) Hammersley low-discrepancy sequence initialization improves the spatial coverage uniformity of the initial population. Compared with pseudo-random initialization, it effectively avoids the search blind zone caused by the initial individual aggregation, enabling the algorithm to discover potential search regions more quickly in the early stage of iteration and accelerate the convergence process.

[0016] (2) Adaptive weighted covariance matrix estimation constructs a covariance matrix with fitness ranking as weight, which makes the construction of the decorrelation space more accurately reflect the advantageous direction of the search space. Compared with the standard covariance estimation in the original BCCO that assigns the same weight to all individuals, it enhances the search efficiency and direction guidance capability of the algorithm in the decorrelation space.

[0017] (3) The elite-guided Cauchy segment exchange mutation strategy injects high-quality gene segments from elite individuals into the population, while using the thick-tailed characteristics of the Cauchy distribution to perform probabilistic bit flip perturbation. The segment length and mutation probability both decrease adaptively with iteration, which accelerates convergence while maintaining population diversity and improves the algorithm's ability to perform local fine search in high-dimensional binary search space.

[0018] (4) Experimental results show that, in 14 fault scenarios of DG distribution networks including single-point faults, multi-point faults and information distortion faults, the average positioning accuracy of the method of the present invention reaches 99.86%, and the positioning accuracy reaches or approaches 100% in all 14 scenarios. It is significantly improved compared with the standard BCCO (95.00%), BGWO (67.71%), BWOA (20.86%) and BPSO (5.43%), demonstrating excellent robustness and practicality. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a distribution network topology containing distributed generation sources. Figure 2 This is an overall flowchart of the method of the present invention; Figure 3 A comparison of the convergence curves of each algorithm in a typical single-point fault scenario (fault segment L11); Figure 4 A comparison of the convergence curves of each algorithm in a typical multi-point fault scenario (fault segment L8+L19+L25); Figure 5 The graph shows a comparison of the convergence curves of each algorithm in a typical distortion fault scenario (fault section L9+L20+L31). Detailed Implementation

[0020] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0021] Example I. Distribution Network Parameter Settings This embodiment uses an IEEE 33-node distribution network with three distributed generation sources (DG1, DG2, and DG3) as an example, and its topology is as follows: Figure 1As shown. This distribution network comprises 33 line sections, numbered L1 to L33, with 33 FTU switch nodes. Based on the distribution network topology, three correlation matrices are pre-established: downstream switch correlation matrix... Upstream switch correlation matrix Association matrix with DG .

[0022] The fault location problem is modeled as a 33-dimensional binary optimization problem. Decision variables. middle, Indicates the line A malfunction occurred. This indicates normal operation; the FTU actually reports the information vector. middle, This indicates that an overcurrent has been detected. This indicates that no overcurrent was detected. This indicates that a reverse fault current has been detected; DG switching state vector middle, express Grid-connected operation This indicates that the user has left the network.

[0023] II. Construction of Fitness Function For a given fault hypothesis vector The fitness function is calculated according to the following procedure: (1) Calculate the downstream fault status of each line. If the line If itself or any of its downstream lines are assumed to be faulty, then Otherwise, it is 0; (2) Calculate the upstream fault status :like Then all lines ; (3) Calculate the DG association state If the line If there is a grid-connected DG among the associated DGs, then ; (4) Generate the desired FTU state: (5) Calculate the fitness value: The first term measures the deviation between the expected state and the actually reported information, while the second term is a penalty to prevent excessive failure assumptions. When hour This is the theoretically optimal value.

[0024] III. IBCCO Algorithm Parameter Settings In this embodiment, the specific parameter settings for the IBCCO algorithm are as follows: population size Maximum number of iterations ; Decision variable dimensions Sigmoid slope parameter Binarization threshold Initial space allocation coefficients , scope of restriction The initial value coefficient for the elite fragment length is 0.3, and the decay coefficient is 0.25; the base value for the Cauchy mutation probability is 0.15.

[0025] IV. Overall Algorithm Flow This embodiment performs fault location and solution according to steps S1 to S9 in the invention description. The overall process is as follows: Figure 2 As shown, specifically: Step S1: Input distribution network parameters, topology matrix, and FTU reported information vector. DG switching state vector ,as well as , , Establish the fitness function. .

[0026] Step S2: Call the Hammersley sequence generation function to generate... Low-discrepancy sequence matrix After linear mapping Zoom to search space Inside.

[0027] Step S3: For each individual Map and use Binarization is performed, the fitness of each individual is calculated, and the global optimal solution is recorded. and its fitness .

[0028] Step S4: Sort the population in ascending order of fitness and initialize. 15 individuals were randomly selected and assigned to the relevant space update group. The remaining 15 were assigned to the original space update group.

[0029] Step S5: By ranking weight Calculate the weighted mean and weighted covariance matrix Perform eigenvalue decomposition Obtain the transformation matrix Transform the population to a derelated space .

[0030] Step S6: For Individuals in a group undergo a center collision strategy position update in the relevant space and are then returned to the original space via an inverse transformation; for the remaining individuals, the center collision strategy position update is directly performed in the original space. The binary mask threshold is used for this purpose. It adapts and changes with iteration.

[0031] Step S7: Perform boundary correction and Sigmoid binary conversion on the updated individuals, calculate fitness, and use a greedy selection algorithm to retain the better solution. Count the number of successes for both groups. , ,according to Adaptive adjustment and limitation According to the new Re-divide into two groups.

[0032] Step S8: After sorting the population by fitness, perform elite-guided Cauchy segment exchange mutation on each individual: according to... from Extract gene fragments to cover the current individual; for uncovered dimensions, ... Combining Cauchy random numbers Perform probabilistic bit flips; greedily select and update the population and the global optimal solution.

[0033] Step S9: Determine if the condition has been met. If the target is not reached, return to step S4 and continue iterating; if the target is reached, output the globally optimal binary solution. The position index with a value of 1 corresponds to the faulty line section number.

[0034] V. Experimental Verification To verify the effectiveness of the method of this invention, simulation experiments were conducted on 14 fault scenarios in distribution networks with distributed generation (DG), including 4 single-point fault scenarios, 4 multi-point fault scenarios, and 6 information distortion fault scenarios. Each algorithm was run independently 50 times in each scenario. Comparison algorithms included: standard BCCO, BGWO, BWOA, and BPSO. The population size for all algorithms was set to 30, the maximum number of iterations was set to 100, and all algorithms used the same Sigmoid binaryization scheme (…). , To ensure a fair comparison, the positioning accuracy of each algorithm is shown in Table 1.

[0035] Table 1. Localization accuracy (%) of each algorithm in 14 fault scenarios As shown in Table 1, the average positioning accuracy of the IBCCO algorithm in this embodiment reaches 99.86% under 14 fault scenarios, which is better than BCCO (95.00%), BGWO (67.71%), BWOA (20.86%) and BPSO (5.43%), especially showing advantages in complex scenarios such as multi-point faults and information distortion faults.

[0036] From the convergence curve (see...) Figure 3 , Figure 4 , Figure 5 These correspond to typical single-point fault scenarios (L11), multi-point fault scenarios (L8+L19+L25), and distortion fault scenarios (L9+L20+L31), respectively). IBCCO exhibits the fastest convergence speed and the best final fitness value in each scenario, typically converging to the global optimum within 20 to 40 iterations. Although BCCO can also reach a relatively good solution in the end, its convergence speed is significantly slower than that of IBCCO. BGWO, BWOA, and BPSO exhibit varying degrees of premature convergence, and their final fitness values ​​are significantly worse than those of IBCCO.

[0037] The experimental results above show that the IBCCO algorithm proposed in this invention can effectively improve the solution accuracy of the center collision optimization algorithm in the distribution network fault location problem through the synergistic effect of three improvements: Hammersley low-difference sequence initialization, adaptive weighted covariance matrix estimation, and elite-guided Cauchy segment exchange mutation.

[0038] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for fault location in a distribution network based on an improved center collision optimization algorithm, characterized in that, Includes the following steps: Step S1: Input distribution network parameters, topology matrix, FTU reporting information, DG switching status, population size, maximum number of iterations, and decision variable dimensions; Based on the distribution network topology, the upstream and downstream correlation matrix of line sections and FTU switches, as well as the DG correlation matrix, are established. The fault location problem is modeled as a binary combinatorial optimization problem with the fault state vector of the line section as the decision variable. A fitness function is constructed with the goal of minimizing the deviation between the expected FTU state and the actual reported FTU information. Step S2: Use Hammersley low-discrepancy sequences to generate an initial population, and scale the sequence points to the feasible region of the decision variables through linear mapping to serve as continuous position values ​​for the initial population; Step S3: Map the initial population using the Sigmoid transfer function and binarize it with a threshold to obtain a binary solution vector. Calculate the fitness value of each individual and record the global optimal solution. Step S4: Sort the population in ascending order according to fitness value, initialize spatial allocation coefficient, and divide the population into two groups: the decorrelated spatial group and the original spatial group. Step S5: Construct a ranking weight vector based on the current population fitness ranking, calculate the weighted mean and weighted covariance matrix, perform eigenvalue decomposition on the weighted covariance matrix to obtain the transformation matrix, and transform the population to the decorrelation space; Step S6: For individuals in the derelated space group, perform a center collision strategy in the derelated space to update their positions and then return to the original space via inverse transformation; for individuals in the original space group, directly perform a center collision strategy in the original space to update their positions. Step S7: Perform boundary correction and Sigmoid binary conversion on all individuals after position update, calculate fitness value and use greedy selection to retain better solutions; count the number of times better offspring are successfully generated in the two space groups respectively, and adaptively update the space allocation coefficients accordingly and limit them to a preset range; Step S8: For each sorted individual, extract a continuous gene fragment from the global optimal solution with a length that decreases with each iteration and cover the corresponding position of the current individual; for the uncovered dimensions, use Cauchy random numbers to perform bit-flip perturbation with a probability that decreases with each iteration; for the mutated individuals, use a greedy strategy to retain the better solution and update the global optimal solution. Step S9: Determine whether the maximum number of iterations has been reached. If not, return to step S4 to continue iterating. If the maximum number of iterations has been reached, output the global optimal binary solution vector, where the line segment corresponding to the position with a value of 1 is the fault location result.

2. The distribution network fault location method based on the improved center collision optimization algorithm according to claim 1, characterized in that, The Hammersley low-difference sequence described in step S2 is generated by sampling the first dimension of the sequence using a uniform distribution. The remaining dimensions use Van der Corput sequences based on prime bases, for the ... Dimension selection prime number , to integer Expand as Number base and fractional reversal: in for exist The digits in the number system; the generated number Quasi-random matrix via linear mapping Zoom to search space , which serves as the continuous position value of the initial population.

3. The distribution network fault location method based on the improved center collision optimization algorithm according to claim 1, characterized in that, The specific method for adaptive weighted covariance matrix estimation in step S5 is as follows: After sorting the population in ascending order of fitness value, the first... The ranking weight of each individual is: in To determine the population size, individuals with higher fitness rankings are given greater weight; a ranking-weight vector is used. Calculate the weighted mean With weighted covariance matrix: Eigenvalue decomposition of the weighted covariance matrix Obtain the transformation matrix Transform the population to a derelated space .

4. The distribution network fault location method based on the improved center collision optimization algorithm according to claim 1, characterized in that, The specific implementation method of the elite-guided Cauchy segment exchange mutation in step S8 is as follows: (4a) Calculate the length of the elite fragment that decreases with iteration: Randomly select the starting position of the segment and end position The global optimal solution exist Gene fragments within the range cover the corresponding locations of the current individual; (4b) Dimensions not covered by elite fragments Calculate the Cauchy mutation probability that decreases with iteration: Generate Cauchy distribution random numbers ,like Then perform a bit flip on that dimension. The multiplier of 10 is used to adjust the effective flip probability of the Cauchy distribution to be equal to... Scale matching; (4c) Calculate the fitness of individuals after mutation. If the result is better than before the mutation, the mutation result is accepted and the global optimal solution is updated; otherwise, the original individual remains unchanged.

Images