A power distribution network fault locating method, device and equipment

Abstract

The application discloses a power distribution network fault positioning method, device and equipment, and belongs to the technical field of intelligent power grid fault positioning. The method comprises the following steps: collecting electrical quantity data to obtain an input data set; adopting a preset multi-dimensional discrete coding mechanism; generating a discrete particle initial population matched with the coding dimension by using chaotic mapping; synchronously and adaptively updating DPSO algorithm parameters, GOA algorithm parameters and common disturbance parameters based on the current iteration progress; obtaining the particle position of the current iteration in the current round iteration; executing a hierarchical escape strategy matched with the current convergence state; executing population diversity screening based on Hamming distance every preset iteration interval; calculating the fitness value of each particle; decoding the final global optimal position into corresponding fault sections, fault positions and fault types, and outputting the fault positioning result. According to the technical scheme, the fault positioning precision can be obviously improved, the convergence speed can be accelerated, and the scheme is suitable for multiple complex distribution network fault scenes.

Description

Technical Field

[0001] This application belongs to the field of intelligent fault location technology for power grids, and specifically relates to a method, device and equipment for fault location in distribution networks. Background Technology

[0002] Fault location in distribution networks is a critical aspect of power operation and maintenance. The industry generally relies on Feeder Terminal Units (FTUs) to collect line electrical quantities and combines this with intelligent optimization algorithms to complete fault identification. Existing mainstream algorithms include Discrete Particle Swarm Optimization (DPSO), Grasshopper Optimization Algorithm (GOA), and simplified combinations of the two. Traditional DPSO uses a fixed learning factor and linear inertia weights, and cannot adaptively switch search priorities throughout the iteration process, making it prone to premature convergence. The original Grasshopper Optimization Algorithm is a continuous numerical optimization architecture, and its encoding form is incompatible with the discrete segment solution scenario of distribution networks. Meanwhile, existing algorithms rely solely on fitness changes as a single criterion to determine local optima, failing to distinguish between different pathological conditions such as population aggregation and particle oscillation. They also lack population diversity control methods adapted to multi-order discrete coding, exhibit poor fault tolerance in the face of field conditions such as FTU data distortion and missing data, low positioning accuracy in high-resistance grounding and multiple fault scenarios, low convergence efficiency, and prominent fault omission issues, making it difficult to adapt to the needs of complex distribution network projects. Summary of the Invention

[0003] This application provides a method, device, and equipment for fault location in distribution networks, aiming to solve the problems of incompatibility of existing DPSO and GOA hybrid algorithms, sudden parameter switching, easy getting trapped in local optima, and poor anti-interference of FTU data. Through a unified framework and hierarchical optimization, the fault location accuracy is significantly improved, the convergence speed is accelerated, and it is suitable for various complex distribution network fault scenarios.

[0004] In a first aspect, embodiments of this application provide a method for locating faults in a power distribution network, the method comprising: S1, Based on the electrical quantity data collected by the FTU measurement points pre-deployed in the distribution network, the electrical quantity data is subjected to distortion correction and preprocessing to obtain the input dataset; S2, based on a preset multi-dimensional discrete coding mechanism, maps each distribution network segment to an independent coding dimension. The coding value is an integer discrete value within a preset interval, which is used to characterize the probability level of a fault occurring in the corresponding segment. S3, using chaotic mapping to generate an initial population of discrete particles that matches the encoded dimension; S4. Based on the current iteration progress, synchronously and adaptively update the DPSO algorithm parameters, GOA algorithm parameters, and common perturbation parameters; the common perturbation parameters are used to uniformly control the random search intensity and escape probability in subsequent iterations. S5. In the current iteration, two sets of candidate position vectors are calculated by adapting the DPSO update rule and GOA update rule to the discrete encoding respectively. The fusion weights of GOA and DPSO are dynamically allocated in combination with the iteration progress and the population diversity state. The two sets of candidate position vectors are weighted and fused, and a chaotic perturbation term based on the common perturbation parameter is superimposed to obtain the particle position of this iteration. S6. Calculate the three-dimensional judgment index based on the particle position in this round. When the preset composite escape condition is met, execute a hierarchical escape strategy that matches the current convergence state. The execution intensity of the hierarchical escape strategy is related to the common perturbation parameter, including: if it is a diversity loss type stagnation, reinitialize some particles; if it is a convergence oscillation type stagnation, perform chaotic mutation on some particles. S7, perform population diversity screening based on Hamming distance at preset iteration intervals, and when the similarity between particles meets the preset conditions, perform chaotic mutation based on the common perturbation parameter on similar particles; S8. Calculate the fitness value of each particle based on the pre-constructed multi-objective fitness function. The fitness value is used to characterize the degree of matching between the fault location result corresponding to the particle and the FTU measured data. Update the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness value, and use the updated global optimal position as the optimization guide for the next iteration. S9, determine whether the current iteration count has reached the preset maximum iteration count: if not, return to step S4 to start the next iteration; if yes, decode the final global optimal position into the corresponding fault segment, fault location and fault type, and output the fault location result.

[0005] Secondly, embodiments of this application provide a power distribution network fault location device, the device comprising: The input dataset construction module is used to collect electrical quantity data based on FTU measurement points pre-deployed in the distribution network, perform distortion correction and preprocessing on the electrical quantity data, and obtain the input dataset. The multidimensional discrete coding module is used to map each distribution network segment to an independent coding dimension based on a preset multidimensional discrete coding mechanism. The coding value is an integer discrete value within a preset interval, which is used to characterize the probability level of a fault occurring in the corresponding segment. The particle swarm initialization module is used to generate an initial population of discrete particles that matches the encoded dimension using chaotic mapping; The iterative parameter synchronization update module is used to synchronously and adaptively update the DPSO algorithm parameters, GOA algorithm parameters, and common perturbation parameters based on the current iteration progress; the common perturbation parameters are used to uniformly control the random search intensity and escape probability in subsequent iterations. The particle position determination module is used to calculate two sets of candidate position vectors in the current iteration by adapting the DPSO update rule and the GOA update rule to the discrete encoding, and dynamically allocate the fusion weights of GOA and DPSO in combination with the iteration progress and the population diversity state. The two sets of candidate position vectors are weighted and fused, and a chaotic perturbation term based on the common perturbation parameter is superimposed to obtain the particle position of the current iteration. The escape control module is used to calculate a three-dimensional judgment index based on the particle position in the current round. When the preset composite escape condition is met, it executes a hierarchical escape strategy that matches the current convergence state. The execution intensity of the hierarchical escape strategy is related to the common perturbation parameter, including: if it is a diversity loss type stagnation, then some particles are reinitialized; if it is a convergence oscillation type stagnation, then chaotic mutation is performed on some particles. The chaotic mutation module is used to perform population diversity screening based on Hamming distance at preset iteration intervals. When the similarity between particles meets the preset conditions, chaotic mutation based on the common perturbation parameter is performed on similar particles. The global optimal position update module is used to calculate the fitness value of each particle based on a pre-constructed multi-objective fitness function. The fitness value is used to characterize the degree of matching between the fault location result corresponding to the particle and the FTU measured data. The module updates the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness value, and uses the updated global optimal position as the optimization guide for the next iteration. The iterative loop control module is used to determine whether the current iteration count has reached the preset maximum iteration count. If not, the next iteration is started. If so, the final global optimal position is decoded into the corresponding fault segment, fault location and fault type, and the fault location result is output.

[0006] Thirdly, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0007] Fourthly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0008] Fifthly, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0009] The technical solution provided in this application significantly improves the algorithm's convergence efficiency through the synergistic effect of unified discrete coding, synchronous parameter control, three-dimensional escape, and Hamming diversity maintenance mechanism. It has high and stable fault location accuracy in distribution networks, can effectively cope with distortion conditions such as noise and data loss, and significantly enhances the ability to identify high-resistance grounding and multiple low-probability faults. Its engineering practicality, especially the feasibility of the solution, has been greatly improved. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating the power distribution network fault location method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the specific process provided in this embodiment; Figure 3 This is a schematic diagram comparing the convergence performance of the four algorithms provided in this embodiment; Figure 4 This is a schematic diagram of the structure of the power distribution network fault location device provided in the embodiments of this application; Figure 5 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0011] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subroutine, etc.

[0012] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0013] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0014] The following description, in conjunction with the accompanying drawings, details the power distribution network fault location method, apparatus, and equipment provided in this application through specific embodiments and application scenarios.

[0015] Example 1 Figure 1 This is a flowchart illustrating the power distribution network fault location method provided in an embodiment of this application. Figure 1 As shown, the specific steps include the following: S1, Based on the electrical quantity data collected by the FTU measurement points pre-deployed in the distribution network, the electrical quantity data is subjected to distortion correction and preprocessing to obtain the input dataset; The power distribution network refers to the 10kV medium-voltage line network used for power transmission in urban and rural areas. Specifically, it can be a radial distribution network topology with a main line of 10km and three T-shaped branches.

[0016] FTU measurement points refer to the data acquisition devices of the Feeder Terminal Unit, which can be set up at substation outlets, line sectionalizing switches, and the beginning of T-junction branches, for a total of 8 monitoring points.

[0017] Electrical quantity data refers to the three-phase voltage, three-phase current, zero-sequence current, switch opening and closing status, and fault alarm signals collected in real time by the FTU equipment.

[0018] Distortion correction refers to the compensation and repair of nine types of abnormal data in the field, such as current deviation, random noise, data loss, and communication interruption.

[0019] Preprocessing refers to the standardized and regularized operations such as time sequence alignment, segment label matching, and numerical normalization.

[0020] The input dataset refers to a structured data set that has been corrected for distortion and standardized in format, making it available for intelligent optimization algorithms to read.

[0021] This solution can continuously collect real-time electrical parameters of the line online through all FTU hardware devices and transmit them to the background analysis server. It can identify the distortion type corresponding to each data point, call the correction algorithm to complete the compensation, and then unify the data dimensions and time series format to generate a standardized input dataset.

[0022] S2, based on a preset multi-dimensional discrete coding mechanism, maps each distribution network segment to an independent coding dimension. The coding value is an integer discrete value within a preset interval, which is used to characterize the probability level of a fault occurring in the corresponding segment. Multidimensional discrete coding mechanism refers to an integer coding rule that uses each line segment as an independent solution dimension. For example, if a distribution network is divided into 8 fault sections, it corresponds to an 8-dimensional coding vector.

[0023] A distribution network section refers to the line section between two adjacent FTU monitoring points, such as the 0km-2km main line section corresponding to FTU 0 to FTU 1.

[0024] The coding dimension refers to the vector solution dimension specific to a single fault section; a line segment corresponds to a one-dimensional code.

[0025] Integer discrete values ​​refer to integers that have no decimals within a defined range and can only take integer values. They are used to quantify the probability of a fault occurring.

[0026] The fault probability level refers to the level of risk of a fault occurring in a given section, with a higher value indicating a higher probability of failure.

[0027] S3, using chaotic mapping to generate an initial population of discrete particles that matches the encoded dimension; Chaotic mapping refers to a mathematical model for generating random sequences that possess the property of uniform spatial traversal.

[0028] The initial population of discrete particles refers to the set of initial solution vectors whose values ​​in all dimensions are integers within a range, which serves as the starting point for the algorithm's iterative search.

[0029] Specifically, the iterative chaotic calculation formula outputs random values, which are then mapped and converted into interval integer particle vectors. The number of vector dimensions of the generated particles is completely consistent with the fault coding dimension.

[0030] S4. Based on the current iteration progress, synchronously and adaptively update the DPSO algorithm parameters, GOA algorithm parameters, and common perturbation parameters; the common perturbation parameters are used to uniformly control the random search intensity and escape probability in subsequent iterations. Iteration progress refers to the percentage value obtained by dividing the current iteration number by the maximum iteration number.

[0031] DPSO algorithm parameters can refer to particle velocity control parameters such as inertia weight, cognitive coefficient, and social coefficient.

[0032] The parameters of the GOA algorithm can refer to grasshopper interaction control parameters such as distance coefficient, attraction strength, and convergence factor.

[0033] The common disturbance parameter refers to the globally unified chaotic disturbance probability, which controls the trigger frequency of particle mutation and local escape throughout the process.

[0034] Synchronous adaptive update refers to adjusting all parameters synchronously based on the same iteration progress, without two independent update logics.

[0035] In this scheme, all particle perturbations and local escape operations within the iteration read the same perturbation probability value.

[0036] S5. In the current iteration, two sets of candidate position vectors are calculated by adapting the DPSO update rule and GOA update rule to the discrete encoding respectively. The fusion weights of GOA and DPSO are dynamically allocated in combination with the iteration progress and the population diversity state. The two sets of candidate position vectors are weighted and fused, and a chaotic perturbation term based on the common perturbation parameter is superimposed to obtain the particle position of this iteration. Here, "iteration round" refers to a single complete parameter update and particle optimization loop. "Update rule" refers to the particle position iterative calculation logic adapted to the integer range of 0 to 10. "Candidate position vector" refers to the integer solution vector output independently by DPSO and GOA. "Population diversity" refers to the degree of numerical difference between particles, quantized and calculated using Hamming distance.

[0037] Fusion weight refers to the weighted proportion of the output results of the two algorithms. Weighted fusion means that the basic position is obtained by weighting and summing the two sets of vector values ​​according to their respective weights.

[0038] The chaotic perturbation term refers to the small, random shift of particles generated based on the common perturbation probability. Dynamic allocation refers to adjusting the weights in real time according to the iteration stage and the population richness. Adding the chaotic shift to the fusion vector generates the final particle position for this round.

[0039] S6. Calculate the three-dimensional convergence criterion based on the particle position in this round. When the preset composite escape condition is met, execute a hierarchical escape strategy that matches the current convergence state. The execution strength of the hierarchical escape strategy is related to the common perturbation parameter, including: if it is a diversity loss type stagnation, reinitialize some particles; if it is a convergence oscillation type stagnation, perform chaotic mutation on some particles. The three-dimensional convergence criterion refers to three sets of judgment data: the magnitude of fitness change, the variance of population distribution, and the proportion of stable particles.

[0040] The composite escape condition refers to the simultaneous fulfillment of the necessary condition of fitness stagnation, combined with any auxiliary condition of population aggregation or particle oscillation.

[0041] The hierarchical escape strategy refers to designing differentiated particle correction schemes for two different ill-conditioning algorithms.

[0042] Diversity loss stagnation refers to an optimization state in which a large number of particles are highly concentrated and the population differences are extremely low.

[0043] Convergent oscillatory stagnation refers to a situation where a particle oscillates back and forth near a local optimum and cannot converge stably.

[0044] Reinitialization refers to selecting a specified proportion of particles and regenerating a completely new integer vector through chaotic mapping.

[0045] Specifically, chaotic sequences can be used to randomly modify the encoded values ​​of each dimension of the particle, and the random amplitude and trigger frequency of the escape operation are uniformly controlled by common perturbation parameters.

[0046] S7, perform population diversity screening based on Hamming distance at preset iteration intervals, and when the similarity between particles meets the preset conditions, perform chaotic mutation based on the common perturbation parameter on similar particles; The iteration interval refers to the fixed number of iteration rounds. In this scheme, the screening is performed once every 5, 8, or 10 generations.

[0047] Hamming distance is a quantitative index of particle similarity obtained by summing the absolute values ​​of the differences between two particles in each dimension.

[0048] Population diversity screening refers to traversing all particle pairs to identify highly repetitive redundant particles.

[0049] Particle similarity refers to the Hamming distance between particles being lower than a preset threshold, indicating that the fault solutions are highly similar.

[0050] Chaotic mutation refers to the random rewriting of the encoding of similar particles based on the probability of a common disturbance.

[0051] Maintaining population diversity means eliminating duplicate particles and expanding the search space of the algorithm.

[0052] S8. Calculate the fitness value of each particle based on the pre-constructed multi-objective fitness function. The fitness value is used to characterize the degree of matching between the fault location result corresponding to the particle and the FTU measured data. Update the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness value, and use the updated global optimal position as the optimization guide for the next iteration. A multi-objective fitness function is a comprehensive evaluation formula that integrates multiple types of penalties and rewards. The fitness value is a numerical value that quantifies the quality of the match between particle estimation faults and real FTU data.

[0053] The individual optimal position refers to the encoding vector with the highest fitness since the beginning of a single particle's iterations. The global optimal position refers to the optimal fault solution vector across all iterations of the entire particle swarm.

[0054] The degree of matching refers to the degree of agreement between the fault section output by the algorithm and the actual fault information measured on site.

[0055] The optimization-oriented approach means that in the next iteration, all particles use the globally optimal vector as the search reference.

[0056] S9, determine whether the current iteration count has reached the preset maximum iteration count: if not, return to step S4 to start the next iteration; if yes, decode the final global optimal position into the corresponding fault segment, fault location and fault type, and output the fault location result.

[0057] The maximum number of iterations refers to the preset upper limit of the algorithm's iterations, which is set to 100 times in this embodiment.

[0058] Fault location results refer to the text report output by the system, which includes the fault section, fault distance, and fault type.

[0059] This technical solution unifies DPSO, GO encoding, and parameter update logic. It overcomes the premature convergence and parameter mutation defects of traditional algorithms through 3D escape and Hamming diversity screening. It is compatible with various FTU data distortion conditions, significantly improves high-resistance and multi-fault identification capabilities, and substantially enhances positioning accuracy. In particular, the convergence speed is significantly improved, avoiding the technical challenge of computer engineering implementation caused by convergence failure. Simultaneously, the field engineering anti-interference capability is greatly enhanced.

[0060] In one embodiment, optionally, in step S2, the discrete integer values ​​within the preset interval are integers between 0 and 10, the encoded values ​​are monotonically positively correlated with the actual fault probability, and the integer encoding can be mapped to the physical fault location on the corresponding segment through a preset decoding rule; In step S3, a chaotic mapping is used to generate an initial population of discrete particles that matches the encoding dimension. The chaotic mapping is a Logistic chaotic mapping, which is used to ensure the uniformity of the distribution of the initial population.

[0061] Integers between 0 and 10 refer to the 11 integers whose encoding dimension values ​​are limited to 0, 1...10, and there are no decimals.

[0062] Monotonic positive correlation means that the larger the coded number, the higher the probability of a fault occurring in the corresponding section.

[0063] Uniform distribution means that the generated initial particles will not be concentrated in a local area, but will cover the entire fault search range.

[0064] This technical solution uses a dedicated, refined integer encoding from 0 to 10 to distinguish it from traditional binary encoding, thereby improving the resolution of fault probability identification. It also establishes a standard Logistic chaotic mapping to avoid the problem of particle clustering in ordinary random initialization. Furthermore, it provides the basic conversion logic for converting the encoding to the actual fault location, thus providing basic conversion support for accurate fault ranging.

[0065] In one embodiment, optionally, in step S2, the decoding rule is: ; in, For the k-th dimension encoded value, The total length of the k-th fault segment; In step S3, the expression for the Logistic chaotic mapping is: ; The generated chaotic sequence is normalized and rounded down to be mapped to an integer code between 0 and 10.

[0066] pos[k] refers to the integer encoded value corresponding to the k-th dimension of the fault segment.

[0067] L k , refers to the total line length of the k-th FTU segment, in km.

[0068] Normalization refers to the computational operation of scaling the chaotic output [0,1] value to the range of 0-10.

[0069] This technical solution provides a precise decoding mathematical formula and a standard Logistic iterative expression, transforming abstract coding into calculable line fault mileage. It fully defines the computational process for converting chaotic sequences into discrete codes, unifies the underlying computational standards of the algorithm, reduces the difficulty of parameter debugging during engineering reproduction, and ensures the consistency of algorithm reproduction under different devices.

[0070] In one embodiment, optionally, in step S4, the synchronous adaptive update uses the iteration progress as a unified benchmark to coordinately adjust the nonlinear inertia weight of DPSO, dynamic learning factor, convergence factor of GOA, distance coefficient, and common perturbation parameter.

[0071] Nonlinear inertia weight refers to the DPSO global search control parameter that changes with the square term of the iteration progress.

[0072] Dynamic learning factors can be divided into cognitive coefficients and social coefficients, which control the learning weights of individuals and groups, respectively.

[0073] The convergence factor is a core parameter of GOA that controls the overall movement step size of the grasshopper.

[0074] The distance factor is a GOA parameter that controls the decay rate of the grasshopper's force.

[0075] This technical solution clearly defines the categories of all controlled parameters within the unified update framework, distinguishes between DPSO-specific parameters, GOA-specific parameters, and global common perturbation parameters, establishes a single-variable collaborative control architecture, and solves the inherent defects of traditional hybrid algorithms where two sets of parameters iterate independently and switch abruptly, ensuring a smooth transition of search intensity during the iteration process.

[0076] In one embodiment, optionally, in step S4, the synchronous adaptive update specifically includes the following formula: DPSO nonlinear inertial weights: ; DPSO cognitive coefficient: ; DPSO social coefficient: ; GOA convergence factor: ; GOA distance coefficient: ; Probability of common disturbance: ; in, .

[0077] `progress` is the iteration progress ratio, serving as the sole independent variable in all formulas. The nonlinear decreasing inertia weight relies on squared terms to achieve rapid decline in the early stages and gradual convergence in the later stages. The cognitive coefficient `c1` is large in the early stages of iteration, emphasizing the particle's own historical failure experience. The social coefficient `c2` is large in the later stages of iteration, emphasizing the global optimal reference of the population. The GOA convergence factor gradually decreases in iterations, reducing the global movement amplitude of grasshoppers in the later stages. The distance coefficient gradually decreases in iterations, weakening the interaction force between grasshoppers at long distances. The common perturbation probability decreases linearly in iterations, resulting in high perturbation in the early stages and stable convergence in the later stages.

[0078] This technical solution provides a complete set of mathematical expressions for synchronous parameter updates, precisely defines the iterative change curves of each parameter, forms standardized parameter linkage control rules, and naturally balances the algorithm's early global exploration and later local development capabilities, significantly reducing the workload of parameter tuning caused by multi-parameter coupling and improving the algorithm's versatility.

[0079] In one embodiment, optionally, in step S5, the dynamic allocation of the fusion weights of GOA and DPSO includes: In the early stage of iteration, DPSO weight is dominant to achieve rapid convergence. In the middle stage of iteration, GOA weight is increased to enhance global search and multi-fault identification capabilities. In the later stage of iteration, DPSO weight is dominant to achieve fine localization. When a decrease in population diversity or convergence stagnation is detected, the GOA weight is adaptively increased to enhance global escape capability.

[0080] The iteration process is divided into three phases: Early iteration (0-30% progress) focuses on quickly identifying the general area of ​​the fault; Mid-iteration (30-70% progress) expands the search scope to investigate multiple faults; Late iteration (70-100% progress) refines and corrects the fault location.

[0081] Fast convergence relies on DPSO local search to quickly narrow down the fault candidate range. Global search relies on GOA for extensive traversal to avoid missing multiple faults. Convergence stagnation is an optimization state where fitness has not significantly improved for several consecutive generations. Adaptive improvement automatically increases the weighting of GOA to enhance global escape capability.

[0082] This technical solution defines the core logic of weight control in three stages, distinguishes the optimization objectives of different iteration stages, and adds an emergency adjustment mechanism for stagnation. Compared with the fixed weight solution, it takes into account the three requirements of rapid locking, multi-fault troubleshooting, and accurate positioning, effectively reducing the probability of missed detection in high-resistance and multi-fault scenarios.

[0083] In one embodiment, optionally, in step S5, both the DPSO update rule and the GOA update rule adopt an improved Sigmoid probabilistic discrete update method, introducing an adaptive temperature parameter: ; Map the continuous update quantity v to the update probability: ; And based on the updated probability, perform discrete jumps of +1 or -1 on the integer encoding; The dynamic allocation of fusion weights for GOA and DPSO specifically includes: When progress < 30%, the DPSO weight is 0.7 and the GOA weight is 0.3. When 30% ≤ progress < 70%, the GOA weight increases linearly from 0.6 to 0.8; When progress ≥ 70%, the DPSO weight is 0.8 and the GOA weight is 0.2. When the change in global optimal fitness over 10 consecutive generations is less than 1e-5, the GOA weight is forcibly increased to 0.9.

[0084] The improved Sigmoid function introduces an adaptive temperature probability transformation function. The adaptive temperature parameter T decreases iteratively, and the initial probability interval is relatively flat, which is beneficial for exploration. The function continuously updates the quantity v, DPSO velocity, and GOA particle interaction force calculation results.

[0085] Discrete jumps are allowed, with only ±1 allowed for the encoded value, and strict constraints on the 0-10 boundary.

[0086] This technical solution unifies the dual-algorithm discrete update mathematical model, provides a temperature adaptive probability conversion formula, accurately solidifies the three-stage weight values ​​and stagnation emergency threshold, eliminates the differences between GOA and DPSO update mechanisms, makes the algorithm fusion and connection smoother, and significantly enhances the global escape capability in the stagnation state.

[0087] In one embodiment, optionally, in step S6, if the change in global optimal fitness over 10 consecutive generations is less than 1e-5 and the variance of the population spatial distribution is less than 1.0, or if the change in global optimal fitness over 10 consecutive generations is less than 1e-5 and more than 60% of the particles have no effective change in position over 3 consecutive generations, then the preset composite escape condition is determined to be met. The graded escape strategy includes: if the stagnation is of the diversity loss type, 20% of the particles are reinitialized; if the stagnation is of the convergent oscillation type, chaotic mutation is performed on 30% of the particles. In step S7, the Hamming distance is calculated using the following formula: ; When the average Hamming distance of the population is less than 8.0, 30% of the particles in similar particle pairs with a Hamming distance of less than 4.0 are subjected to chaotic mutation, while the top 20% of elite particles in terms of fitness are retained. In step S8, the multi-objective fitness function includes one or more of the following: fault segment matching penalty, location accuracy segmentation penalty, FTU distortion penalty, load rationality penalty, and cable type matching reward. The weight of each penalty is adapted according to the actual probability of the distribution network fault. The weight of a single-phase ground fault is 1.0, and the weight of a high-resistance ground fault is 0.65.

[0088] Among them, the change in global optimal fitness is the numerical difference in evaluation between adjacent optimal solutions. Population spatial distribution variance is a statistical indicator measuring the degree of particle aggregation.

[0089] Elite particles, those with high fitness rankings and excellent fault-solving abilities, are directly retained and do not participate in mutation.

[0090] The average Hamming distance and the mean similarity of all particles reflect the overall population richness.

[0091] A multi-objective fitness function and a multi-dimensional comprehensive scoring model that takes into account accuracy, distortion, and fault type.

[0092] This technical solution fully integrates three-dimensional judgment thresholds, graded escape ratios, Hamming distance screening criteria, and multi-target fitness sub-rules, closely matching the characteristics of real distribution network fault distribution. It can accurately distinguish different local optimal types and handle them accordingly, demonstrating extremely high positioning accuracy under data distortion and high-resistance grounding conditions.

[0093] To enable those skilled in the art to better understand this solution, this application also provides a preferred embodiment.

[0094] This embodiment uses a typical distribution network topology of 10kV, 10km long main distribution line + 3 T-connection branches as the simulation object to fully implement the fault location method of the present invention. The specific content is as follows: 1. Configuration of FTU measurement points in the distribution network; A total of 8 FTU measurement points were set up, with the following specific configuration: -FTU0: Installed at the 0km end of the main line (substation outlet), it collects three-phase voltage, three-phase current, zero-sequence current, switch status, and fault alarm signals on the power supply side at the beginning of the line, corresponding to the monitoring section of the main line from 0 to 2km. -FTU1: Installed at 2km on the main line (before the T1 branch connection), it collects all electrical and status quantities at the sectionalizing switch on the main line, corresponding to the monitoring of the 2-5km section of the main line; -FTU1-1: Installed at the switch at the beginning of branch T1, it collects three-phase current, zero-sequence current, and fault alarm signals, and monitors the entire section of branch T1 accordingly; -FTU2: Installed at 5km of the main line (before the T2 branch junction), it collects all electrical and status quantities, corresponding to the monitoring section of the main line from 5-8km; -FTU2-1: Installed at the switch at the beginning of the T2 branch, it collects core current and alarm signals, and monitors the entire section of the T2 branch accordingly; -FTU3: Installed at 8km of the main line (before the T3 branch junction), it collects all electrical and status quantities, corresponding to the monitoring section of the main line from 8 to 10km. -FTU3-1: Installed at the switch at the beginning of the T3 branch, it collects core current and alarm signals, and monitors the entire section of the T3 branch accordingly; -FTU4: Installed at the end switch of the main line 10km away, it collects all electrical and status quantities for end boundary verification.

[0095] 2. Fault type and weight configuration; The weights are set according to the actual fault probability of the distribution network, specifically: single-phase grounding weight 1.0, two-phase short circuit weight 0.85, two-phase grounding weight 0.85, three-phase short circuit weight 0.7, high-resistance grounding weight 0.65, connection point overheating weight 0.5, and insulator pollution flashover weight 0.4.

[0096] 3. Algorithm core parameter settings; The population size is set to 50, the maximum number of iterations is set to 100, the particle dimension is consistent with the number of faulty sections in the distribution network, and the encoding range is 0-10 integers; DPSO parameters: W_MAX=0.9, W_MIN=0.4, C1_MAX=2.5, C1_MIN=1.5, C2_MAX=1.5, C2_MIN=0.5; GOA parameters: distance coefficient range 1.5-2.5, attraction strength range 0.5-1.0, goa_c_max=1.0, goa_c_min=0.00001; chaotic perturbation probability range 0.1-0.4; three-dimensional decision thresholds: stagnation threshold 1e-5, diversity variance threshold 1.0, particle no change ratio threshold 60%; Hamming distance screening threshold 8.0, screening is performed every 5 generations.

[0097] It should be emphasized that the three-dimensional judgment parameters used in this scheme include a stagnation threshold of 1e-5, a diversity variance threshold of 1.0, and a particle no-change ratio of 60%. The combination of these three judgment conditions is an innovative design of this scheme.

[0098] In addition, using Hamming distance for population diversity maintenance is an effective improvement strategy in this scheme.

[0099] This solution not only improves the construction of a unified adaptive parameter framework, but also improves the code's _update_parameters_unified( The method achieves unified parameter updates for GOA and DPSO, uses a three-dimensional joint decision escape mechanism, and employs dynamic master-slave weight allocation to dynamically adjust the GOA / DPSO weights based on the iteration stage and population diversity.

[0100] This is particularly difficult to achieve in terms of improvement, requiring not only parameter coordination to ensure that GOA and DPSO parameters work together, but also the design of an escape mechanism to balance diversity and convergence speed. Furthermore, the weight allocation strategy is not as simple as conventional weight settings; it needs to be tailored to the model's performance, dynamically adjusting the master-slave relationship.

[0101] 4. Complete fault location execution process; Step 1: Chaotic initialization: The initial particle swarm is generated using Logistic chaotic mapping, with each dimension mapped to an integer from 0 to 10, thus completing the population initialization; Step 2: Iteration starts, with the current iteration number starting from 1. Calculate the iteration progress: progress = iteration / max_iter; Step 3: Unified parameter update: Call the _update_parameters_unified method to synchronously update the inertia weight w, learning factors c1 / c2, GOA core parameters, chaotic perturbation probability, and GOA / DPSO mixed weights; Step 4: Three-dimensional joint escape determination: Call the _check_escape_condition method to determine whether the composite condition of "stagnation + loss of diversity / convergent oscillation" is met; if it is met, if diversity is lost, reinitialize 20% of the particles, and if convergent oscillation occurs, perform chaotic mutation on 30% of the particles; if it is not met, proceed to the next step. Step 5: GOA position update: Calculate the interaction force between grasshoppers, convert the force into position update probability through the improved adaptive temperature Sigmoid function, and complete the position update based on the 0-10 discrete encoding based on the probability; Step 6: DPSO position update: Update particle velocity based on dynamic learning factor and nonlinear inertia weight, with velocity constrained in the range of [-3,3], and complete discrete position update through Sigmoid function; Step 7: Hybrid Position Fusion: According to the adaptively assigned GOA / DPSO weights, the update results of the two algorithms are weighted and fused. The formula is: positions = dpso_weight × dpso_positions + goa_weight × goa_positions. Step 8: Diversity screening: Hamming distance screening is performed every 5 generations. The Hamming distance of all particle pairs is calculated. Chaotic mutation is performed on particles with similarity exceeding the threshold to maintain population diversity. Step 9: Fitness Calculation: Collect real-time FTU data. If distortion exists, call the _apply_ftu_distortion method to process it and calculate the multi-objective fitness value, including fault segment penalty, position accuracy segment penalty, FTU distortion penalty, load rationality penalty, and cable type reward. Step 10: Optimal value update: Update the individual optimal position and the global optimal position, and record the iteration history data; Step 11: Termination judgment: If the iteration reaches the maximum number of iterations, terminate the iteration and output the fault segment and fault location corresponding to the global optimal position; if it does not reach the maximum number of iterations, increment the iteration by 1 and return to step 2 to continue the iteration.

[0102] 5. Verify the effect with examples; In this example, the positioning accuracy reached over 96% for scenarios such as single-phase grounding, two-phase short circuit, high-resistance grounding, and multiple faults. Even with 15% random noise in the FTU, 200ms data delay, and some data loss, it can still achieve stable and accurate positioning, reducing the number of convergence iterations by more than 40% compared to traditional algorithms.

[0103] The simulation results for positioning accuracy are shown in the table below:

[0104] Based on the above examples, it can be seen that the core improvements of this solution include, but are not limited to, the following: 1. A GOA-DPSO hybrid algorithm framework based on 0-10 unified integer discrete coding, and a GOA-Sigmoid probabilistic discrete position update method that matches the DPSO mechanism; 2. An adaptive improvement method for DPSO based on dynamic learning factors and nonlinear inertia weights, specifically a dynamic adjustment rule for c1 decreasing from 2.5 to 1.5 and c2 increasing from 0.5 to 1.5, and a nonlinear inertia weight calculation method for w=0.9-0.5×progress²; 3. A dynamic master-slave weight allocation method for GOA and DPSO based on iterative state and Hamming distance population diversity, including phased weight allocation rules and a forced weight adjustment mechanism when convergence stalls; 4. A three-dimensional joint judgment rule of "fitness stagnation + population diversity loss + particle convergence oscillation", and a hierarchical local optimal escape strategy for the corresponding diversity loss restart and convergence oscillation mutation. 5. A Hamming distance-based population diversity maintenance method adapted to 0-10 discrete encoding, including a complete process of Hamming distance difference quantification, similar particle detection, and periodic screening of chaotic mutations; 6. Nine types of distortion processing methods adapted to the FTU acquisition scenario of power distribution network, and a method for constructing a multi-objective fault location fitness function with FTU distortion robustness penalty term; 7. A weighted adaptation mechanism based on the actual occurrence probability of distribution network faults, and a corresponding method for optimizing fault location accuracy.

[0105] The technical difficulty of implementing a weighted adaptation mechanism for the actual probability of distribution network faults, compared to conventional weight configuration, can be seen in the table below:

[0106] For the step-by-step implementation process of the chaotic GOA-DPSO hybrid algorithm, Figure 2 This is a schematic diagram of the specific process provided in this embodiment. For example... Figure 2 As shown, the main steps include the following: Step 1: 0-10 Integer Multidimensional Discrete Encoding Scheme; Implementation logic: Fault parameters (location, resistance, type) are encoded as integers in the range of 0-10, with each dimension representing a candidate fault segment and the value indicating the probability level of a fault occurring in that segment. This encoding method abandons traditional binary encoding, refines the search space, and significantly improves positioning resolution.

[0107] It is important to emphasize here that this solution does not involve conventional parameter settings and level classifications. Specifically, those skilled in the art will be unable to overcome the following practical technical problems: Barrier 1: Algorithm adaptation barrier; The essence of the problem: DPSO and GOA were originally designed for continuous optimization, and their direct application to discrete space will cause problems.

[0108] The velocity-position model is not applicable: the velocity of a traditional PSO is continuous, but our position is an integer from 0 to 10, which cannot be directly added together; The force vector problem in GOA: The force calculated by GOA is a continuous vector, which cannot be directly applied to discrete positions; The updated rules require a redesign: changing from "acceleration" to "based on probability increase or decrease of 1" requires a completely new mathematical model; The key challenge is to design a Sigmoid probabilistic mechanism to convert continuous force / velocity into discrete position increments.

[0109] Barrier Two: Adaptability Design Barrier; The essence of the problem: Traditional fitness functions are designed based on "whether a fault exists or not", while probabilistic coding needs to consider "the probability of a fault".

[0110] The traditional approach has failed: the original fitness function only cared about whether the faulty section was correctly identified; Multi-objective trade-offs: It is necessary to consider both positioning accuracy and the rationality of probability distribution; The key challenge is establishing a mapping relationship between probability and actual failures, and designing a reasonable weighting mechanism.

[0111] Barrier Three: Parameter Tuning Barrier; The essence of the problem: the new encoding method introduces a large number of new parameters, and there are complex interactions between these parameters.

[0112] The number of parameters has exploded: from the traditional 3-5 parameters to more than 15; Coupling effect: Adjusting the inertia weight may affect the effect of the learning factor, and adjusting the escape threshold may change the effect of chaotic perturbation; Lack of theoretical guidance: Most parameters are empirical values, and there are no general formulas that can be directly applied; High computational cost: Each parameter combination requires running a complete optimization process to verify the effect; The key challenge is to design a systematic parameter sensitivity analysis to identify key parameters and robust ranges.

[0113] Barrier Four: Engineering Implementation Barriers; The essence of the problem: Discrete coding has many error-prone details at the code implementation level.

[0114] Boundary handling: The position must be strictly limited to the range of 0-10; exceeding the range will result in a logical error. Data type: Integer type must be used; floating-point errors will cause unexpected behavior. Distance metric: Traditional Euclidean distance is no longer applicable, and Hamming distance needs to be used instead; Chaotic mapping adaptation: Chaotic sequences need to be correctly mapped to the range of 0-10; Overcoming the challenges: It is necessary to handle a large number of boundary cases and special scenarios to ensure the robustness of the algorithm.

[0115] Barrier Five: Verification and Evaluation Barrier; The essence of the problem: new methods need to be proven to be better than traditional methods, which requires rigorous experimental design and statistical analysis.

[0116] Benchmark selection: A suitable comparison algorithm needs to be selected to ensure fairness; Indicator Design: Traditional positioning accuracy indicators are no longer fully applicable, and new evaluation indicators need to be designed. Statistical significance: It is necessary to prove through statistical tests that the performance improvement is not accidental; Reproducibility: Other researchers need to be able to reproduce the result; Overcoming the challenges: A comprehensive experimental plan needs to be designed, covering different scenarios and parameter configurations.

[0117] Specifically, the mathematical principle of this scheme is as follows: Encoding space: pos[i]∈{0,1,2,3,4,5,6,7,8,9,10} (discrete integers); Encoding dimensions: num_dimensions=10 (multidimensional encoding); Decoding formula: real_location = pos[0] / 10 × line_length (mapped to actual position); Core code: ChaoticGOADPSO.py, lines 74-77; 0-10 integer multidimensional discrete encoding; def_chaotic_initialization(self): Initialize particle positions (discrete values, integers in the range of 0-10) using a chaotic sequence. positions=[]; for_inrange(self.num_particles): position=[]; for_inrange(self.num_dimensions): self.chaotic_value=self._logistic_map(self.chaotic_value); Mapped to the discrete coding range of 0-10; position.append(int(self.chaotic_value×10)) is a 0-10 integer encoding; positions.append(position); return np.array(positions); ChaoticGOADPSO.py, lines 439-455; def update_position(self): Update particle position (discrete integer); Use an improved Sigmoid function to convert velocity into probability; sigmoid=self._improved_sigmoid (self.velocities); Position is updated based on probability (keeping the discrete range of 0-10); foriinrange(self.num_particles): forjinrange(self.num_dimensions): if np.random.random() <sigmoid[i,j]: Add position (discrete incremental update); self.positions[i,j]=min(10,self.positions[i,j]+1); else: Reduce the number of positions; self.positions[i,j]=max(0,self.positions[i,j]-1); Step 2: Logistic chaos initialization; Implementation logic: The Logistic chaotic mapping is used to generate chaotic sequences. The ergodicity of chaotic motion is used to ensure that the initial particles are uniformly distributed in the search space, thus avoiding the uneven distribution problem that may be caused by traditional random initialization.

[0118] Mathematical principles: Logistic chaotic mapping ; When μ=4, the system is in a completely chaotic state, and the sequence traverses the interval [0,1]. Initial value x(0) Random selection ensures diversity in each run; Mapping formula: pos = round(x × 10) → {0, 1, 2, ..., 10}; Core code: ChaoticGOADPSO.py, lines 313-315; Logistic chaotic mapping; def _logistic_map(self, x): Logistic chaotic mapping: x = 4 × x × (1 - x); return 4 × x × (1 - x); # ChaoticGOADPSO.py Lines 317-326; def _chaotic_initialization(self): Chaotic initialization: The ergodicity of the chaotic sequence is used to ensure a uniform initial distribution; positions = [] for _ in range(self.num_particles): position = [] for _ in range(self.num_dimensions): # Logistic chaotic mapping iteration self.chaotic_value = self._logistic_map(self.chaotic_value) # Mapping to the 0-10 discrete coding range position.append(int(self.chaotic_value × 10)) positions.append(position) return np.array(positions); It should be noted that this is one of the innovative aspects of this solution, and is not a conventional setting. Please refer to the table below for details:

[0119] # Lines 72-73: Chaotic sequence parameter initialization; self.chaotic_value = np.random.random() # Randomly initialize chaotic values ​​Step 3: Unified adaptive parameter update framework; Implementation logic: The GOA and DPSO algorithms share the same set of parameter update logic, including inertia weights, learning factors, GOA core parameters, and chaotic perturbation probabilities. The progress variable enables synchronous adaptive adjustment of all parameters, ensuring smooth switching between the two algorithms.

[0120] Mathematical Principles [Inertia Weight - Non-linear Decreasing] w(t) = w_max - (w_max - w_min) × (t / T)² = 0.9 - (0.9 - 0.4) × progress² In the early stages, a higher weight is allocated to ensure exploration; in the later stages, a lower weight is allocated to ensure development. [Cognitive coefficient c1 - linearly decreasing] c1(t) = c1_max - (c1_max - c1_min) × progress = 2.5 - (2.5 - 1.5) × progress The early stage focuses on individual cognition, while the later stage focuses on social learning; [Social coefficient c² - linearly increasing] c2(t) = c2_min + (c2_max - c2_min) × progress = 0.5 + (1.5 - 0.5) × progress;

GOA Parameters - Adaptive Adjustment

[0121] Traditional hybrid algorithms suffer from parameter discontinuity during switching: when switching from DPSO to GOA, parameters may change abruptly, causing a sudden change in search behavior and affecting convergence stability.

[0122] The unified parameter update framework proposed in this paper ensures parameter continuity and search behavior consistency during algorithm switching, which is an effect that traditional methods cannot achieve.

[0123] The key point here is that all parameters are adjusted synchronously based on the same progress variable. Ensure consistent and seamless parameter update logic during GOA and DPSO switching. progress = iteration / self.max_iter # Progress variable [0,1]; Nonlinear inertia weights (used in DPSO) In the early stages, a higher weight is used to ensure exploration; in the later stages, a lower weight is used to ensure development. self.w = self.W_MAX - (self.W_MAX - self.W_MIN) × (progress 2 ); Dynamic learning factor (used by DPSO) The early stage focuses on cognition (C1 University), while the later stage focuses on society (C2 University). self.c1 = self.C1_MAX - (self.C1_MAX - self.C1_MIN) × progress# 2.5→1.5 self.c2 = self.C2_MIN + (self.C2_MAX - self.C2_MIN) × progress# 0.5→1.5; GOA parameters (used by GOA) As the distance coefficient increases, the attraction intensity decreases. self.goa_c = self.goa_c_max - (self.goa_c_max - self.goa_c_min) ×progress self.goa_distance_coeff = 1.5 + (2.5 - 1.5) × (1 - progress) self.goa_attraction_strength = 0.5 + 0.5 × (1 - progress) Chaotic perturbation probability (shared by both) High perturbations in the early stage enhance exploration, while low perturbations in the later stage ensure convergence. self.chaotic_prob = self.CHAOTIC_PROB_MAX - \ (self.CHAOTIC_PROB_MAX - self.CHAOTIC_PROB_MIN) × progress It should be noted that the problem with the traditional approach is that the two sets of parameters may not be compatible. When w=0.5 for DPSO, c for GOA may be 0.3. The search strengths of the two are inconsistent, and the two sets of parameters evolve independently, which may lead to "strong-strong conflict" or "weak-weak superposition". The parameters are discontinuous when switching algorithms, resulting in sudden changes in search behavior. Parameter tuning is difficult and both sets of parameters need to be adjusted simultaneously.

[0124] Innovative approach: Unified parameter update framework: def_update_parameters_unified(self,iteration): progress = iteration / self.max_iter # Single progress variable #All parameters are adjusted synchronously based on the same progress. self.w=self.W_MAX-(self.W_MAX-self.W_MIN)× (progress 2 ) self.c1=self.C1_MAX-(self.C1_MAX-self.C1_MIN)×progress self.c2=self.C2_MIN+(self.C2_MAX-self.C2_MIN)×progress self.goa_c=self.goa_c_max-(self.goa_c_max-self.goa_c_min)×progress self.goa_distance_coeff=1.5+ (2.5-1.5)×(1-progress) self.goa_attraction_strength=0.5+0.5×(1-progress) self.chaotic_prob=self.CHAOTIC_PROB_MAX-(self.CHAOTIC_PROB_MAX-self.CHAOTIC_PROB_MIN)×progress Core of Innovation: Single progress variable: All parameters are adjusted synchronously based on the same progress. Parameter consistency: Ensure that the search intensity of GOA and DPSO is coordinated at any time; Seamless switching: Parameters change continuously without abrupt changes when switching algorithms; I. Framework Design Concept: The core idea is to synchronously update all adaptive parameters of both GOA and DPSO algorithms based on the same progress variable, ensuring continuous and seamless parameter switching during algorithm transitions.

[0125] Design goals: Parameter coordination: Avoiding parameter conflicts between two algorithms; Seamless switching: ensuring continuity of search behavior; Simplified parameter tuning: One set of parameters controls two algorithms; II. In-depth analysis of the core code; 1. Framework Entry Point Key Design: Single progress variable: progress = current iteration count / maximum iteration count Range: From 0 (start of iteration) to 1 (end of iteration) Function: To serve as a unified benchmark for all parameter updates; 2. DPSO parameter update: Mathematical formula:

[0126] Parameter configuration: W_MAX = 0.9 (initial iteration) W_MIN = 0.4 (End of iteration) Design principles: Non-linear decreasing: using squared terms, the weight decreases rapidly in the early stages and slowly in the later stages; Exploration phase (progress < 0.5): w ≈ 0.7-0.9, maintaining significant inertia to promote global exploration; During the development phase (progress>0.5): w ≈ 0.4-0.7, reduce inertia and enhance local search; Mathematical formula: ; ; Parameter configuration: C1_MAX = 2.5, C1_MIN = 1.5; C2_MAX = 1.5, C2_MIN = 0.5; Design principles: The cognitive coefficient c1 decreases: individual experience is emphasized in the early stages and weakened in the later stages; The social coefficient c² increases: group experience is underestimated in the early stages and strengthened in the later stages; Balancing mechanism: Early exploration (individual-oriented) → Later development (group-oriented); 3. GOA parameter update: Mathematical formula: ; ; ; Parameter configuration: goa_c_max = 1.0, goa_c_min = 0.00001; distance_coeff: 1.5 (late stage) → 2.5 (early stage); attraction_strength: 0.5 (late stage) → 1.0 (early stage); Design principles:

[0127] 4. Chaotic perturbation probability update; Chaotic perturbation probability (shared by both), high perturbation probability promotes exploration, while low perturbation probability ensures stability in the later stages. self.chaotic_prob = self.CHAOTIC_PROB_MAX - (self.CHAOTIC_PROB_MAX - self.CHAOTIC_PROB_MIN) × progress; Mathematical formula: ; Parameter configuration: CHAOTIC_PROB_MAX = 0.4 (initial iteration); CHAOTIC_PROB_MIN = 0.1 (end of iteration); Design principles: Early high perturbation (40% probability): breaks local optima and promotes exploration; Low perturbation in the later stage (10% probability): Maintains the stability of the solution and ensures convergence; 5. Adaptive weight allocation (core of hybrid strategy); Python adaptive weight allocation (hybrid strategy) dynamically adjusts GOA / DPSO weights based on population diversity. diversity = self._calculate_avg_hamming_distance(); diversity_ratio = min(diversity / self.MAX_DIVERSITY, 1.0); # Enhance Global Search (GOA) when diversity is high, enhance Local Search (DPSO) when diversity is low. self.dpso_weight = 0.5 + 0.3 × (1 - diversity_ratio) ; self.goa_weight = 1 - self.dpso_weight; # If convergence stalls, increase the GOA weights to perform a global exploration; if self._check_stagnation(): self.goa_weight = min(0.9, self.goa_weight + 0.2) self.dpso_weight = 1 - self.goa_weight;

[0128]

[0129] Step 4: Three-dimensional joint judgment escape mechanism; Implementation logic: The algorithm employs a three-dimensional composite decision rule: "fitness stagnation + population diversity loss + particle convergence oscillation." An escape strategy is triggered only when this composite condition is met, preventing the algorithm from getting trapped in local optima. The necessary condition must be met, and the algorithm is triggered when at least one auxiliary condition is satisfied.

[0130] The three-dimensional joint decision-making escape mechanism employed here is not a conventional design and is the innovation of this solution, for the following reasons: 1: Expansion of the three-dimensional judgment dimension; Innovative content: Expanding from one-dimensional / two-dimensional judgment to three-dimensional judgment

[0131] Innovation Value: Three-dimensional judgment can more comprehensively diagnose the algorithm state. Different dimensions correspond to different types of convergence problems, providing a basis for hierarchical escape strategies.

[0132] 2: Design of Composite Decision Logic Innovative content: A compound logic of "necessary conditions + auxiliary conditions" was designed. escape_triggered = condition1 and (condition2 or condition3)

[0133] First, stagnation is a necessary condition: if fitness is still improving, escape should not occur even if diversity is low or there are oscillations (it may just be normal convergence). Second, the two auxiliary conditions are "or": loss of diversity and convergence oscillations are two different pathological conditions, and either condition can trigger escape. Furthermore, it's not a matter of "all three conditions being met": overly strict conditions can lead to insufficient escape, preventing timely escape from local optima.

[0134] 3: Matching of graded escape strategies; Innovative Feature: Matching different escape strategies based on trigger conditions.

[0135] Technical barrier 1: The essential differences between the three pathological conditions are difficult to identify;

[0136] 1. The three pathological conditions may present with similar symptoms (all manifesting as "fitness no longer improves"); 2. In-depth analysis of the algorithm's internal states is required to distinguish them; 3. Most researchers only focus on "stagnation" and ignore other pathological conditions.

[0137] Technical barrier 2: The design of composite decision logic requires in-depth understanding; 1. It is necessary to understand that "stagnation" is a prerequisite for escape (the algorithm must first get stuck in a dilemma). 2. It is necessary to understand that "loss of diversity" and "oscillation" are two different pathologies (which require different treatments). 3. A reasonable logical combination needs to be designed to avoid being too strict or too lenient.

[0138] Technical barrier 3: The matching of graded escape strategies requires experimental verification;

[0139] 1. Extensive experiments are needed to determine the restart / mutation ratio; 2. The effectiveness of strategy matching needs to be verified; 3. It needs to be ensured that escape does not destroy the search results.

[0140] Technical barrier 4: Coordinated optimization of parameter thresholds

[0141] 1. The five parameters interact with each other, forming a "parameter coupling network"; 2. Different optimization problems require completely different parameter settings; 3. There is a lack of universal theoretical guidance, requiring a large number of experiments.

[0142] In existing literature, escape mechanisms are mostly one-dimensional (stagnation only) or two-dimensional (stagnation + diversity), and three-dimensional mechanisms are not yet available.

[0143] This paper introduces "convergence oscillation judgment" as a third dimension for the first time, which is a novel extension to the diagnosis of algorithmic pathologies. Convergence oscillation and loss of diversity are two fundamentally different pathologies: Diversity loss: population clustering restricts search space; Convergent oscillation: The particle oscillates near the local optimum and cannot converge stably; Identifying these two pathological conditions requires completely different detection methods.

[0144] In addition, the composite judgment logic in this paper is not a simple "all conditions are satisfied" or "any condition is satisfied", but an innovative logic of "necessary conditions + auxiliary conditions" is designed. The theoretical basis for this logical design is: Stagnation is a necessary condition: if fitness is still improving, there is no need to escape; Loss of diversity and oscillation are auxiliary conditions: two different pathologies, either of which is sufficient. This logic design avoids the problems of being "too strict" (insufficient escape) and "too lenient" (excessive escape), and requires in-depth theoretical analysis and experimental verification.

[0145] Finally, the graded escape strategy proposed in this paper is not simply "different strategies for different situations," but rather a precise matching of pathological types and treatment methods. This precise matching requires: 1. Gain a deep understanding of the essential differences between the two pathological conditions; 2. Design targeted processing strategies; 3. The effectiveness of the matching was verified through a large number of experiments.

[0146] Mathematical principles:

Necessary Condition - Stagnation Detection

Auxiliary Condition 1 - Determination of Diversity Loss

Auxiliary Condition 2 - Convergence Oscillation Judgment

Triggering Rules

[0147] Mathematical principles: [Weight Allocation During Iteration Phase] Early stage (0-30%): DPSO=70%, GOA=30% → Fast convergence Mid-game (30-70%): GOA = 60%-80% → Global Exploration Later stage (70-100%): DPSO=80%, GOA=20% → Fine positioning [Dynamic Adjustment Based on Diversity] diversity_ratio = min(avg_hamming_distance / MAX_DIVERSITY, 1.0) When diversity is low → increase GOA weight to enhance global search. When diversity is high, increase DPSO weight to strengthen local development. Forced adjustments during convergence stagnation When convergence rate < 0.01, force GOA weight = 0.9 and DPSO weight = 0.1.

Comprehensive Weighting Formula

[0148] Mathematical principles: Definition of Hamming distance Hamming(pos1, pos2) = Σ|pos1[i] - pos2[i]| = |pos1[0]-pos2[0]| + |pos1[1]-pos2[1]| + ... For 0-10 encoding, the range is: [0, num_dimensions × 10] Average Hamming Distance avg_hamming = (1 / C_n^2) × Σ Σ Hamming(pos_i, pos_j) Where C_n^2 = n(n-1) / 2 is the number of particle pairs. [Similar Particle Determination] if avg_hamming <threshold (threshold=8.0): Find all particle pairs where Hamming(pos_i, pos_j) < threshold / 2 Perform chaotic mutation on 30% of similar particle pairs

Diversity screening formula

[0149] 1. Traditional Hamming distance (only applicable to binary encoding); The problem with the traditional binary Hamming distance is that it is only applicable to 0 / 1 encoding, and cannot distinguish the difference between "difference of 1" and "difference of 10" in 0-10 encoding.

[0150] 2. Euclidean distance (applicable to continuous encoding); The problem with Euclidean distance is that it is applicable to continuous values, but may generate non-integer distances for integer encoding, does not consider the discrete nature of encoding, and for 0-10 encoding, the range of Euclidean distance is not intuitive.

[0151] 3. Improved Hamming distance definition; Calculate the Hamming distance between two particle positions (applicable to 0-10 discrete encoding). The Hamming distance is the sum of the absolute values of the differences between corresponding dimensions of two vectors. For 0-10 integer encoding, the range is [0, num_dimensions×10], which can distinguish differences of different degrees. The distance range is directly related to the encoding range, and it is applicable to any integer encoding range.

[0152] Average Hamming distance calculation Design principle: Calculate the average value of the Hamming distances between all particle pairs; Reflects the overall diversity level of the population; A higher numerical value indicates greater diversity; The complexity of threshold setting; Difficulty Analysis: The challenges of setting thresholds: Dimensionality-related: The more dimensions, the larger the maximum Hamming distance; Related to the encoding range: 0-10 encoding and 0-100 encoding require different thresholds; Related to the characteristics of the problem: different optimization problems require different thresholds; Threshold calculation example: Assume there are 10 dimensions, each ranging from 0 to 10. Maximum Hamming distance: 10 × 10 = 100 Average Hamming distance range: [0, 100] A threshold of 8.0 means that the average difference per dimension is less than 0.8. First, there is no universal formula for direct calculation; second, improper threshold setting can affect algorithm performance.

[0153] def _calculate_avg_hamming_distance(self): Calculate the average Hamming distance of the population if self.num_particles<2: return 0.0 total_distance = 0.0 count = 0 for i in range(self.num_particles): for j in range(i + 1, self.num_particles): total_distance += self._hamming_distance(self.positions[i],self.positions[j]) count += 1 return total_distance / count if count>0 else 0.0 ChaoticGOADPSO.py, lines 278-311 def _hamming_diversity_filter(self, threshold=8.0): Hamming distance screening: Detecting and handling overly similar particles This process is repeated every 5 generations to maintain population diversity. avg_hamming = self._calculate_avg_hamming_distance() Sufficient diversity, no need for screening if avg_hamming>= threshold: return 0 Identify overly similar particle pairs (Hamming distance) <threshold / 2) similar_pairs = [] for i in range(self.num_particles): for j in range(i + 1, self.num_particles): distance = self._hamming_distance(self.positions[i], self.positions[j]) if distance <threshold / 2: similar_pairs.append((i, j, distance)) Perform chaotic mutations on similar particles (processing up to 30% of similar pairs). mutated_count = 0 num_to_mutate = int(len(similar_pairs) × 0.3) # 30% threshold for i, j, _ in similar_pairs[:num_to_mutate]: Chaotic mutation of the second particle for dim in range(self.num_dimensions): self.chaotic_value = self._logistic_map(self.chaotic_value) if np.random.random() < 0.3: self.positions[j, dim] = int(self.chaotic_value × 10) mutated_count += 1 return mutated_count; Step 7: Improve the probabilistic update of Sigmoid; Implementation logic: An improved adaptive sigmoid function is used to convert particle velocity into position update probability. A temperature parameter is introduced to achieve a smooth curve in the early stage of iteration (exploration) and a steep curve in the later stage (convergence). Boundary correction is added to avoid the probability from being completely close to 0 or 1.

[0154] Mathematical principles: Standard Sigmoid sigmoid(v) = 1 / (1 + e^(-v)) [Adaptive Temperature Sigmoid] temperature(t) = 2.0 - 1.5 × (t / T_max) sigmoid(v, T) = 1 / (1 + e^(-v / T)) High temperature (T large) → Flat curve → Probability close to 0.5 → High exploratory nature Low temperature (T small) → Steep curve → Probability close to 0 or 1 → High exploitability Boundary Correction sigmoid = clip(sigmoid, 0.01, 0.99) To avoid probabilities being entirely 0 or 1, a certain degree of randomness must be maintained. Location Update Rules if random() <sigmoid(v): pos += 1 (increase)

[0155] else: pos -= 1 (decrease)

[0156] Core code: ChaoticGOADPSO.py, lines 416-437 Improved Sigmoid function: def _improved_sigmoid(self, velocity): Improved Sigmoid function: Adaptive temperature + boundary correction The core of this improvement lies in: achieving high exploration in the early stages and high development in the later stages of temperature parameter iteration. Adaptive temperature: dynamically adjusted according to the iteration stage if hasattr(self, '_current_iteration'): The temperature decreased linearly from 2.0 to 0.5. temperature = 2.0 - (1.5 × self._current_iteration / self.max_iter) else: temperature = 1.0 (default temperature) Temperature-dependent Sigmoid curves: the curve is flatter at high temperatures (exploration), and steeper at low temperatures (convergence). sigmoid = 1 / (1 + np.exp(-velocity / temperature)) Boundary correction: Avoid probabilities that are exactly close to 0 or 1, maintaining a certain degree of randomness. sigmoid = np.clip(sigmoid, 0.01, 0.99) return sigmoid # ChaoticGOADPSO.py Lines 439-455 def update_position(self): Position update using an improved Sigmoid The decision to add or remove positions is based on probability (keeping the discrete range of 0-10). sigmoid = self._improved_sigmoid(self.velocities) for i in range(self.num_particles): for j in range(self.num_dimensions): if np.random.random() <sigmoid[i, j]: Add positions (no more than 10). self.positions[i, j] = min(10, self.positions[i, j]+ 1) else: Reduce positions (not lower than 0). self.positions[i, j] = max(0, self.positions[i, j]- 1); Step 8: GOA / DPSO unified location update blend; Implementation logic: Figure 3 This is a diagram comparing the convergence performance of the four algorithms provided in this embodiment. Figure 3 As shown, GOA and DPSO independently update the candidate positions, and then perform a weighted mixture using dynamically assigned weights. DPSO is responsible for fine-grained local search, while GOA is responsible for global escape exploration; the two complement each other.

[0157] The design philosophy of this scheme is as follows: When the algorithm stagnates in convergence (without significant improvement in fitness over multiple generations), it indicates that the algorithm may be trapped in a local optimum. At this point, it is necessary to forcibly increase the GOA weight to enhance the global exploration capability and help the algorithm escape from the local optimum.

[0158] Dynamic principle: Stagnation detected → Force increase GOA weight (up to 0.9); No stagnation detected → Maintain normal weights; Stagnation detection method: Judgment conditions: Stagnation Decision = (Change in global optimal fitness over T consecutive generations < ε) ; Where T=10 (the threshold for continuous stagnation algebra). ε = 1e-5 (threshold for fitness change). Mandatory adjustment rules: If the stall detection is true: goa_weight=min(0.9,goa_weight+0.2) dpso_weight=1-goa_weight Adjustment effect: Each time stagnation is detected, the GOA weight increases by 0.2. The maximum weight of GOA is 0.9. Ensure that 10% of the DPSO weight is retained even in extreme cases. Mathematical principles: [GOA Location Update] forces[i] = c × Σ s(r_ij) × (pos[j]- pos[i]) / r_ij s(r) = f × exp(-r / l) - exp(-r) (attraction function) Where f = 0.5, l = 1.5 new_pos_GOA = gbest + c × forces → Sigmoid probabilistics → Discrete integers

DPSO location update

[0159] The technical effect achieved by this patent is: 1. Significantly improved positioning accuracy: The positioning accuracy of distribution network faults has been improved from 90%-92% of traditional algorithms to over 96%, which can stably achieve accurate positioning of high-resistance grounding, multiple faults, and low-probability faults, greatly reducing the missed detection rate.

[0160] 2. Significantly faster convergence speed: Through dynamic learning factors, nonlinear inertia weights, and adaptive weight allocation, the algorithm's iterative convergence speed is improved by more than 40% compared to traditional algorithms, and it can quickly locate the candidate fault region.

[0161] 3. Significantly enhanced global optimization and resistance to local optima: Through a three-dimensional joint decision escape mechanism, chaotic perturbation, and Hamming distance diversity maintenance, the algorithm can effectively escape local optima and effectively alleviate the premature convergence problem.

[0162] 4. Significantly improved engineering robustness: Through nine types of FTU distortion processing and distortion penalty mechanisms, the algorithm can still maintain stable positioning performance under harsh conditions such as FTU data noise, loss, distortion, and communication interruption, making it suitable for actual power distribution network engineering scenarios.

[0163] 5. Significantly improved algorithm stability and scalability: Through unified discrete coding, unified parameter update framework, and unified probabilistic update mechanism, the hybrid switching between GOA and DPSO is smooth and seamless, and the unified algorithm framework facilitates subsequent maintenance and expansion.

[0164] Example 2 Figure 4 This is a schematic diagram of the structure of the power distribution network fault location device provided in the embodiments of this application. Figure 4 As shown, the device includes: The input dataset construction module 401 is used to collect electrical quantity data based on FTU measurement points pre-deployed in the distribution network, perform distortion correction and preprocessing on the electrical quantity data, and obtain the input dataset. The multidimensional discrete coding module 402 is used to map each distribution network segment to an independent coding dimension based on a preset multidimensional discrete coding mechanism. The coding value is an integer discrete value within a preset interval, which is used to characterize the probability level of a fault occurring in the corresponding segment. The particle swarm initialization module 403 is used to generate an initial population of discrete particles that matches the encoded dimension using chaotic mapping; The iteration parameter synchronization update module 404 is used to synchronously and adaptively update the DPSO algorithm parameters, GOA algorithm parameters, and common perturbation parameters based on the current iteration progress; the common perturbation parameters are used to uniformly control the random search intensity and escape probability in subsequent iterations. The particle position determination module 405 is used to calculate two sets of candidate position vectors in the current iteration by adapting the DPSO update rule and the GOA update rule to the discrete encoding, and dynamically allocate the fusion weights of GOA and DPSO in combination with the iteration progress and the population diversity state, weightedly fuse the two sets of candidate position vectors, and superimpose the chaotic perturbation term based on the common perturbation parameter to obtain the particle position of the current iteration. The escape control module 406 is used to calculate a three-dimensional judgment index based on the particle position in the current round. When the preset composite escape condition is met, it executes a hierarchical escape strategy that matches the current convergence state. The execution intensity of the hierarchical escape strategy is related to the common perturbation parameter, including: if it is a diversity loss type stagnation, then some particles are reinitialized; if it is a convergence oscillation type stagnation, then chaotic mutation is performed on some particles. The chaotic mutation module 407 is used to perform population diversity screening based on Hamming distance at preset iteration intervals. When the similarity between particles meets the preset conditions, chaotic mutation based on the common perturbation parameter is performed on similar particles. The global optimal position update module 408 is used to calculate the fitness value of each particle based on a pre-constructed multi-objective fitness function. The fitness value is used to characterize the degree of matching between the fault location result corresponding to the particle and the FTU measured data. The module updates the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness value, and uses the updated global optimal position as the optimization guide for the next iteration. The iterative loop control module 409 is used to determine whether the current iteration count has reached the preset maximum iteration count. If not, the next iteration is started. If so, the final global optimal position is decoded into the corresponding fault segment, fault location and fault type, and the fault location result is output.

[0165] The power distribution network fault location device provided in this application can realize the various processes of the above embodiments, and will not be described again here to avoid repetition.

[0166] Example 3 like Figure 5 As shown, this application embodiment also provides an electronic device 500, including a processor 501, a memory 502, and a program or instructions stored in the memory 502 and executable on the processor 501. When the program or instructions are executed by the processor 501, they implement the various processes of the above-described distribution network fault location method embodiment and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0167] It should be noted that the electronic devices in the embodiments of this application include mobile electronic devices and non-mobile electronic devices as described above.

[0168] Example 4 This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described distribution network fault location method embodiments and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0169] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0170] Example 5 This application also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described distribution network fault location method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0171] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system on a chip, etc.

[0172] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element. Furthermore, it should be noted that the scope of the methods and systems in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0173] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0174] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms fall within the scope of protection of this application.

[0175] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A method for locating faults in a power distribution network, characterized in that, The method includes: S1, Based on the electrical quantity data collected by the FTU measurement points pre-deployed in the distribution network, the electrical quantity data is subjected to distortion correction and preprocessing to obtain the input dataset; S2, based on a preset multi-dimensional discrete coding mechanism, maps each distribution network segment to an independent coding dimension. The coding value is an integer discrete value within a preset interval, which is used to characterize the probability level of a fault occurring in the corresponding segment. S3, using chaotic mapping to generate an initial population of discrete particles that matches the encoded dimension; S4. Based on the current iteration progress, synchronously and adaptively update the DPSO algorithm parameters, GOA algorithm parameters, and common perturbation parameters; the common perturbation parameters are used to uniformly control the random search intensity and escape probability in subsequent iterations. S5. In the current iteration, two sets of candidate position vectors are calculated by adapting the DPSO update rule and GOA update rule to the discrete encoding respectively. The fusion weights of GOA and DPSO are dynamically allocated in combination with the iteration progress and the population diversity state. The two sets of candidate position vectors are weighted and fused, and a chaotic perturbation term based on the common perturbation parameter is superimposed to obtain the particle position of this iteration. S6. Calculate the three-dimensional judgment index based on the particle position in this round. When the preset composite escape condition is met, execute a hierarchical escape strategy that matches the current convergence state. The execution intensity of the hierarchical escape strategy is related to the common perturbation parameter, including: if it is a diversity loss type stagnation, reinitialize some particles; if it is a convergence oscillation type stagnation, perform chaotic mutation on some particles. S7, perform population diversity screening based on Hamming distance at preset iteration intervals, and when the similarity between particles meets the preset conditions, perform chaotic mutation based on the common perturbation parameter on similar particles; S8. Calculate the fitness value of each particle based on the pre-constructed multi-objective fitness function. The fitness value is used to characterize the degree of matching between the fault location result corresponding to the particle and the FTU measured data. Update the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness value, and use the updated global optimal position as the optimization guide for the next iteration. S9, determine whether the current iteration count has reached the preset maximum iteration count: if not, return to step S4 to start the next iteration; if yes, decode the final global optimal position into the corresponding fault segment, fault location and fault type, and output the fault location result.

2. The method according to claim 1, characterized in that, In step S2, the discrete integer values ​​within the preset interval are integers between 0 and 10. The encoded value is monotonically positively correlated with the actual fault probability, and the integer encoding can be mapped to the physical fault location on the corresponding segment through preset decoding rules. In step S3, a chaotic mapping is used to generate an initial population of discrete particles that matches the encoding dimension. The chaotic mapping is a Logistic chaotic mapping, which is used to ensure the uniformity of the distribution of the initial population.

3. The method according to claim 2, characterized in that, In step S2, the decoding rule is as follows: ； in, For the k-th dimension encoded value, The total length of the k-th fault segment; In step S3, the expression for the Logistic chaotic mapping is: ; The generated chaotic sequence is normalized and rounded down to be mapped to an integer code between 0 and 10.

4. The method according to claim 1, characterized in that, In step S4, the synchronous adaptive update uses the iteration progress as a unified benchmark to coordinately adjust the nonlinear inertia weight and dynamic learning factor of DPSO, the convergence factor and distance coefficient of GOA, and the common perturbation parameter.

5. The method according to claim 4, characterized in that, In step S4, the synchronous adaptive update specifically includes the following formula: DPSO nonlinear inertial weights: ; DPSO cognitive coefficient: ; DPSO social coefficient: ; GOA convergence factor: ; GOA distance coefficient: ; Probability of common disturbance: ; in, .

6. The method according to claim 1, characterized in that, In step S5, the dynamic allocation of fusion weights for GOA and DPSO includes: In the early stages of iteration, DPSO weights are dominant to achieve rapid convergence; in the middle stages of iteration, GOA weights are increased to enhance global search and multi-fault identification capabilities; and in the later stages of iteration, DPSO weights are dominant to achieve fine localization during regression. When a decrease in population diversity or convergence stagnation is detected, the GOA weight is adaptively increased to enhance global escape capability.

7. The method according to claim 6, characterized in that, In step S5, both the DPSO update rule and the GOA update rule adopt an improved Sigmoid probabilistic discrete update method, introducing an adaptive temperature parameter: ; Map the continuous update quantity v to the update probability: ; And based on the updated probability, perform discrete jumps of +1 or -1 on the integer encoding; The dynamic allocation of fusion weights for GOA and DPSO specifically includes: When progress < 30%, the DPSO weight is 0.7 and the GOA weight is 0.

3. When 30% ≤ progress < 70%, the GOA weight increases linearly from 0.6 to 0.8; When progress ≥ 70%, the DPSO weight is 0.8 and the GOA weight is 0.

2. When the change in global optimal fitness over 10 consecutive generations is less than 1e-5, the GOA weight is forcibly increased to 0.

9.

8. The method according to claim 1, characterized in that, In step S6, if the change in global optimal fitness is less than 1e-5 for 10 consecutive generations and the variance of population spatial distribution is less than 1.0; or if the change in global optimal fitness is less than 1e-5 for 10 consecutive generations and more than 60% of the particles have no effective change in position for 3 consecutive generations, then the preset composite escape condition is satisfied. The graded escape strategy includes: if the stagnation is of the diversity loss type, then 20% of the particles are reinitialized; if the stagnation is of the convergent oscillation type, then 30% of the particles are subjected to chaotic mutation. In step S7, the Hamming distance is calculated using the following formula: ; When the average Hamming distance of the population is less than 8.0, 30% of the particles in similar particle pairs with a Hamming distance of less than 4.0 are subjected to chaotic mutation; at the same time, the top 20% of elite particles in terms of fitness are retained. In step S8, the multi-objective fitness function includes one or more of the following: fault segment matching penalty, location accuracy segmentation penalty, FTU distortion penalty, load rationality penalty, and cable type matching reward. The weight of each penalty is adapted according to the actual probability of the distribution network fault. The weight of a single-phase ground fault is 1.0, and the weight of a high-resistance ground fault is 0.

65.

9. A power distribution network fault location device, characterized in that, The device includes: The input dataset construction module is used to perform distortion correction and preprocessing on the electrical quantity data collected by FTU measurement points pre-deployed in the distribution network to obtain the input dataset; The multidimensional discrete coding module is used to map each distribution network segment to an independent coding dimension based on a preset multidimensional discrete coding mechanism. The coding value is an integer discrete value within a preset interval, which is used to characterize the probability level of a fault occurring in the corresponding segment. The particle swarm initialization module is used to generate an initial population of discrete particles that matches the encoded dimension using chaotic mapping; The iterative parameter synchronization update module is used to synchronously and adaptively update the DPSO algorithm parameters, GOA algorithm parameters, and common perturbation parameters based on the current iteration progress; the common perturbation parameters are used to uniformly control the random search intensity and escape probability in subsequent iterations. The particle position determination module is used to calculate two sets of candidate position vectors in the current iteration by adapting the DPSO update rule and the GOA update rule to the discrete encoding, and dynamically allocate the fusion weights of GOA and DPSO in combination with the iteration progress and the population diversity state. The two sets of candidate position vectors are weighted and fused, and a chaotic perturbation term based on the common perturbation parameter is superimposed to obtain the particle position of the current iteration. The escape control module is used to calculate a three-dimensional judgment index based on the particle position in the current round. When the preset composite escape condition is met, it executes a hierarchical escape strategy that matches the current convergence state. The execution intensity of the hierarchical escape strategy is related to the common perturbation parameter, including: if it is a diversity loss type stagnation, then some particles are reinitialized; if it is a convergence oscillation type stagnation, then chaotic mutation is performed on some particles. The chaotic mutation module is used to perform population diversity screening based on Hamming distance at preset iteration intervals. When the similarity between particles meets the preset conditions, chaotic mutation based on the common perturbation parameter is performed on similar particles. The global optimal position update module is used to calculate the fitness value of each particle based on a pre-constructed multi-objective fitness function. The fitness value is used to characterize the degree of matching between the fault location result corresponding to the particle and the FTU measured data. The module updates the individual optimal position of each particle and the global optimal position of the particle swarm according to the fitness value, and uses the updated global optimal position as the optimization guide for the next iteration. The iterative loop control module is used to determine whether the current iteration count has reached the preset maximum iteration count. If not, the next iteration is started. If so, the final global optimal position is decoded into the corresponding fault segment, fault location and fault type, and the fault location result is output.

10. An electronic device, characterized in that, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the power distribution network fault location method as described in any one of claims 1-8.

Images