Power grid fault traveling wave positioning method based on fusion optimization algorithm

Abstract

The application provides a power grid fault traveling wave positioning method based on a fusion optimization algorithm, comprising: establishing a dynamic virtual fault framework model of a power grid, randomly generating a plurality of virtual fault particles containing position coordinates, and constructing an initial virtual fault particle cluster; using a shortest path algorithm to calculate the topological path length of each virtual fault particle in the virtual fault particle cluster to all measurement points in the power grid, and then calculating a virtual fault space-time feature vector; setting the traveling wave arrival time when a real fault occurs as a target clustering centroid in a feature space, and calculating a feature matching fitness; constructing a non-convex optimization problem with the optimal feature matching fitness as a target, solving the problem using a fusion optimization algorithm, and thus obtaining a fault positioning result and a traveling wave velocity, so as to realize traveling wave positioning. The application can solve the problems of local optimum, slow convergence speed and high calculation cost in the prior art.

Description

Power grid fault traveling wave localization method based on fusion optimization algorithm Technical Field

[0001] This invention relates to the field of power system data processing technology, and more specifically to a method for locating traveling waves of power grid faults based on a fusion optimization algorithm. Background Technology

[0002] The power system is a critical infrastructure for the operation of modern society. With the expansion of the power grid and the large-scale integration of new energy sources, the probability of power grid failures is increasing, and the difficulty of fault location is also increasing. Rapid and accurate fault location technology has become a key link in improving power supply reliability. Existing traveling wave network location methods integrate data from the entire network, which partially overcomes the above limitations, but still need to address challenges such as wave velocity uncertainty, topology changes, and multivariate coupling. With the integration of distributed energy resources and the development of smart grids, the power grid structure is becoming increasingly complex, and fault location needs to balance accuracy, efficiency, and adaptability, driving the continuous exploration of new optimization algorithms.

[0003] To overcome the shortcomings of traditional methods, the traveling wave network localization method constructs a global observation by using the initial arrival times of traveling waves at multiple measurement points. It then uses optimization algorithms to adjust the location and wave velocity of the virtual fault point to match the characteristics of the real fault, thereby achieving localization. However, existing methods are mostly based on deterministic models, such as virtual fault localization techniques based on particle swarm optimization (PSO) and aurora optimization (PLO) algorithms, which are prone to problems such as local optima, slow convergence speed, and high computational cost. Summary of the Invention

[0004] In view of this, the present invention provides a power grid fault traveling wave localization method based on a fusion optimization algorithm to solve the problems of existing technologies that are prone to local optima, slow convergence speed and high computational cost.

[0005] A power grid fault traveling wave localization method based on a fusion optimization algorithm includes:

[0006] Step S1: Establish a dynamic virtual fault framework model of the power grid. In the dynamic virtual fault framework model, randomly generate several virtual fault particles containing position coordinates. Use the virtual fault particles as the initial signal source of the virtual fault traveling wave to construct the initial virtual fault particle cluster.

[0007] Step S2: The shortest path algorithm is used to calculate the topological path length from each virtual fault particle in the virtual fault particle cluster to all measurement points in the power grid. Combined with the preset traveling wave velocity, the theoretical absolute time of the virtual fault traveling wave arriving at each measurement point is calculated. Then, the theoretical absolute time is mapped to the virtual fault spatiotemporal feature vector in the high-dimensional feature space.

[0008] Step S3: Obtain the arrival time of the traveling wave recorded at each measurement point when the real fault occurs, set the arrival time of the traveling wave as the centroid of the target cluster in the feature space, calculate the generalized distance between the virtual fault spatiotemporal feature vector and the centroid of the target cluster, and use the generalized distance as the feature matching fitness that quantifies the virtual fault particles belonging to the real fault category.

[0009] Step S4: Construct a non-convex optimization problem with the goal of achieving optimal feature matching fitness. Solve the problem using a fusion optimization algorithm to obtain the fault location result and traveling wave velocity, thereby achieving traveling wave localization. This fusion optimization algorithm uses a probabilistic surrogate model to fit the objective function. The core of the model adopts a multi-scale feature coupling kernel function. It uses adaptive weighted coupling of local non-smooth components and global isotropic components, and uses an adaptive hybrid acquisition function to dynamically balance global exploration and local development. Furthermore, this fusion optimization algorithm introduces a dual convergence criterion based on expected information gain and feature space compactness to dynamically evaluate the optimization state during the iteration process.

[0010] The power grid fault traveling wave localization method based on fusion optimization algorithm provided by the present invention has the following beneficial effects:

[0011] (1) This invention avoids the rigid virtual / real time difference matrix comparison mode in traditional methods. It innovatively adopts a matching mechanism between the virtual fault spatiotemporal feature vector and the target cluster centroid. This idea of ​​transforming fault location into feature matching fitness optimization in the feature space not only reduces the computational dimension and avoids the storage and computational overhead caused by multi-dimensional matrix operations, but also enhances the method's tolerance to noisy data. Combined with a fusion optimization algorithm based on a probabilistic surrogate model, it can quickly lock the global optimal solution in a non-convex, multi-peak fault search space. Compared with traditional exhaustive or deterministic search algorithms, it accelerates the convergence speed and reduces the computational cost, significantly improving the real-time performance and applicability of the location process, and is especially suitable for rapid fault diagnosis of large-scale power grids.

[0012] (2) Multi-scale accurate fitting of complex traveling wave transmission characteristics is achieved, enhancing the robustness of positioning. This invention introduces a multi-scale feature coupling kernel function into the fusion optimization algorithm, replacing the single kernel function model. By adaptively coupling local non-smooth components and global isotropic components, it can not only capture the macroscopic attenuation trend of long-distance traveling wave transmission (global features), but also accurately characterize the local small disturbances caused by wave velocity uncertainty and line impedance changes (local features). Combined with the adaptive hybrid acquisition function developed by dynamic balance exploration and development, as well as the dual convergence criteria, it can dynamically adjust the search confidence boundary according to the sparsity of the feature space, effectively avoiding falling into the local optimum trap, and achieving high-precision adaptive positioning in complex power grid topology environments. Attached Figure Description

[0013] Figure 1 is a flowchart illustrating the power grid fault traveling wave location method based on fusion optimization algorithm provided in an embodiment of the present invention;

[0014] Figure 2 is an exemplary simulation model topology diagram. Detailed Implementation

[0015] Please refer to Figure 1. The power grid fault traveling wave localization method based on fusion optimization algorithm provided by the present invention includes steps S1 to S4:

[0016] Step S1: Establish a dynamic virtual fault framework model of the power grid. In the dynamic virtual fault framework model, randomly generate several virtual fault particles containing position coordinates. Use the virtual fault particles as the initial signal source of the virtual fault traveling wave to construct the initial virtual fault particle cluster.

[0017] First, a dynamic virtual fault framework model of the power grid is established. This model describes the mathematical model of the power grid topology and line lengths, providing basic physical parameters and path constraints for calculating the propagation distance from all fault points to the measurement points in subsequent steps. Then, within the dynamic virtual fault framework model, several virtual fault particles containing location coordinates are randomly generated. These virtual fault particles serve as the initial signal source for the virtual fault traveling wave, thereby constructing the initial virtual fault particle cluster.

[0018] Step S2: The shortest path algorithm is used to calculate the topological path length from each virtual fault particle in the virtual fault particle cluster to all measurement points in the power grid. Combined with the preset traveling wave velocity, the theoretical absolute time of the virtual fault traveling wave arriving at each measurement point is calculated. Then, the theoretical absolute time is mapped to the virtual fault spatiotemporal feature vector in the high-dimensional feature space.

[0019] Traditional location methods typically rely on absolute time differences, which are easily limited by the time synchronization accuracy of the measuring device. This invention treats the traveling wave signal generated by the virtual fault point as a set of feature sequences containing spatiotemporal information. By introducing feature centering processing, the influence of absolute time on the feature vector is eliminated, thereby constructing a virtual fault spatiotemporal feature vector that is only related to the fault location and wave velocity. This vector represents a virtual particle to be verified in the feature space.

[0020] Step S2 specifically includes:

[0021] The shortest path algorithm is used to calculate the topological path length from each virtual fault particle in the virtual fault particle cluster to all measurement points in the power grid. The expression is as follows:

[0022]

[0023] in, Indicates virtual fault point To the Measurement points The shortest topological path length; This represents the shortest path algorithm; in this embodiment, it is specifically Dijkstra's algorithm. Virtual fault point Location coordinates in the power grid topology.

[0024] Assuming the fault occurs at time 0, and considering the preset traveling wave velocity... The theoretical absolute time for the arrival of the virtual fault traveling wave at each measurement point is calculated using the following expression:

[0025] ;

[0026] in, For the virtual fault traveling wave to reach the first Measurement points The absolute moment of the theory.

[0027] To eliminate interference from the unknown initial moment of a fault, a virtual fault spatiotemporal feature vector is constructed. Unlike traditional pairwise subtraction, this invention employs a mean-centered method, projecting the time-series data onto a zero-mean feature plane. Specifically, the theoretical absolute moment is mapped to a virtual fault spatiotemporal feature vector in a high-dimensional feature space, expressed as:

[0028]

[0029] in, Represents the spatiotemporal feature vector of a virtual fault. , , , They represent The 1st, the 2nd, the 3rd The, the Each feature component The total number of measurement points. Indicates transpose;

[0030] The expression is:

[0031]

[0032]

[0033] in, The temporal centroid of the virtual time series sequence. For the virtual fault traveling wave to reach the first The theoretical absolute time of each measurement point.

[0034] Constructed It is a pure spacetime feature descriptor that does not depend on the origin time.

[0035] Step S3: Obtain the arrival time of the traveling wave recorded at each measurement point when the real fault occurs, set the arrival time of the traveling wave as the centroid of the target cluster in the feature space, calculate the generalized distance between the spatiotemporal feature vector of the virtual fault and the centroid of the target cluster, and use the generalized distance as the feature matching fitness that quantifies the virtual fault particles belonging to the real fault category.

[0036] In this approach, the data collected when real faults occur is defined as the target cluster centroid in the feature space. Essentially, the problem is no longer solving a system of equations, but rather finding the virtual particle whose generalized spatial distance (i.e., feature matching fitness) is closest to the centroid. Processing real data as centroid vectors allows the search process to be transformed into a self-organizing process of particle swarms converging towards the centroid during subsequent fusion optimization, thus satisfying the requirements of the fusion optimization algorithm.

[0037] In this embodiment, step S3 specifically includes:

[0038] When a real fault occurs, define the set of arrival times of the traveling waves recorded at each measurement point. for:

[0039]

[0040] in, , , They are the 1st, 2nd, and 3rd respectively. The arrival time of the traveling wave recorded at each measurement point;

[0041] By using the feature space mapping rule, the arrival time of the traveling wave is set as the centroid of the target cluster in the feature space. The expression is:

[0042]

[0043] in, , , , They are the 1st, 2nd, and 3rd respectively. The, the One centroid component;

[0044] The expression is:

[0045]

[0046]

[0047] in, For the first The arrival time of the traveling wave recorded at each measurement point. For the first The arrival time of the traveling wave recorded at each measurement point. The centroid of the actual measurement data in the time domain;

[0048] The generalized distance between the spatiotemporal feature vector of the virtual fault and the centroid of the target cluster is calculated. This generalized distance is used as the feature matching fitness quantified to characterize the virtual fault particles belonging to the real fault category. The expression is as follows:

[0049]

[0050] in, The feature matching fitness is represented by the weighted Euclidean distance between the virtual fault feature vector and the target cluster centroid, which characterizes the probability that the virtual fault point belongs to the real fault category. To optimize the combination of variables, , For the first The reliability weighting coefficient for each measurement point is used to suppress feature distortion caused by excessive measurement noise or blurred wavefront identification. The default setting is [not specified]. .

[0051] when When the value approaches 0, it indicates that the spatiotemporal characteristic distribution of virtual particles highly overlaps with that of real faults. This is the optimal estimate of the actual fault location.

[0052] By using mean centering, the timing of actual fault occurrences was addressed. This unknown problem does not need to be searched for during the optimization process. This reduces the optimization dimensionality and improves convergence efficiency.

[0053] Step S4: Construct a non-convex optimization problem with the goal of achieving optimal feature matching fitness. Solve the problem using a fusion optimization algorithm to obtain the fault location result and traveling wave velocity, thereby achieving traveling wave localization. This fusion optimization algorithm uses a probabilistic surrogate model to fit the objective function. The core of the model adopts a multi-scale feature coupling kernel function. It uses adaptive weighted coupling of local non-smooth components and global isotropic components, and uses an adaptive hybrid acquisition function to dynamically balance global exploration and local development. Furthermore, this fusion optimization algorithm introduces a dual convergence criterion based on expected information gain and feature space compactness to dynamically evaluate the optimization state during the iteration process.

[0054] Among them, the traditional gradient descent method is prone to getting trapped in local optima, while the surrogate model with a single kernel function is difficult to fit simultaneously the macroscopic attenuation law (low-frequency trend) of traveling wave signal in long-distance transmission and the microscopic random disturbances caused by wave speed uncertainty (high-frequency non-smooth characteristics).

[0055] To address this, this application proposes a fusion optimization algorithm. This algorithm establishes a probabilistic surrogate model based on a Gaussian process, with the core being the design of a multi-scale feature coupling kernel function. This kernel function decouples the response surface of the objective function into a globally isotropic component and a locally non-smooth component. Through adaptive weighted coupling of these two components, high-precision fitting of the fault feature space is achieved. Simultaneously, an adaptive hybrid acquisition strategy is introduced to dynamically balance the exploration of unknown regions and the development of known regions during the iteration process, guiding the search path to quickly approach the global optimum.

[0056] Specifically, the multi-scale feature coupling kernel function is defined as a linear coupling form between local and global features, expressed as:

[0057]

[0058]

[0059]

[0060] in, It is a multi-scale feature coupling kernel function; This is a multi-scale coupling factor used to adjust the weight ratio of local and global features in the surrogate model. ; The component is a globally isotropic component used to capture the macroscopic smooth trend of the fault feature matching fitness as a function of distance (i.e., the path decay law on a large scale). This component assumes that the changes in fault features are continuous and infinitely differentiable in a vast topological space, ensuring the algorithm's ability to macroscopically lock onto faulty regions with a high probability. This is a local non-smooth component used to capture local objective function jitter (i.e., small-scale non-smooth features) caused by line impedance discontinuity or small changes in traveling wave velocity. This component allows the objective function to have finite differentiability locally, thereby accurately fitting the "spur-like" feature surface caused by parameter uncertainty in actual working conditions and preventing the algorithm from failing in the fine search stage. For the Kronecker function, To observe the noise variance, The global squared exponent. For global feature scale parameters, and The first The moment and the The optimal combination of variables at each time step. It is a local squared exponent. For smoothness parameters, It is the Gamma function; The local feature scale parameter is used to measure the correlation distance of local non-smooth features in the feature space. The smaller the value, the stronger the surrogate model's ability to capture high-frequency disturbances. This is a modified Bessel function of the second type.

[0061] Furthermore, in order to select the next optimal sampling point based on the above probabilistic proxy model, this invention designs an adaptive hybrid sampling function. , It combines the advantages of Expected Improvement (EI) and Confidence Upper Bound (UCB), and introduces a dynamic balancing weight, expressed as:

[0062]

[0063]

[0064] in, For dynamic balancing weight function, Represents the mathematical expectation. This indicates taking the maximum value. The fitness of the currently known minimum feature matching. and The Gaussian process is in the th... The mean and standard deviation of the predictions from each iteration. This is the confidence interval adjustment coefficient; The strategy switching inflection point is used to define the iteration moment when the algorithm's focus smoothly transitions from global exploration to local optimization, ensuring that the algorithm has sufficient space traversal capability in the early stage and extremely high convergence accuracy in the later stage. This is a strategy switching rate factor used to adjust the sensitivity of the adaptive hybrid acquisition function during the transition between exploration and exploitation; a larger value indicates a higher sensitivity. The value indicates that the algorithm will reach the switching inflection point. It then quickly converges to a local optimization mode, making it suitable for high signal-to-noise ratio environments; smaller The value makes the weight changes smoother, prolongs the time the algorithm stays in the mixed search state, and thus enhances the robustness of the algorithm in escaping the local optimum trap in a noisy environment.

[0065] In the early stages of the iteration Approaching 0, the algorithm focuses on the UCB (upper confidence bound), tending to explore unknown regions with high uncertainty in the feature space to avoid missing the global optimum; as the number of iterations increases, As the value gradually approaches 1, the algorithm smoothly transitions to EI (Expected Improvement), focusing on fine-tuning near known extreme points to quickly converge to the true fault location.

[0066] To balance location accuracy and computational efficiency in complex power grid fault inversion processes, this invention abandons the traditional single iterative termination strategy and designs a dual convergence criterion combining feature space compactness and expected information gain. The dual convergence criterion based on expected information gain and feature space compactness includes a feature space compactness criterion. And expected information gain criterion .

[0067] Feature space compactness criterion Used to evaluate whether the current virtual fault feature vector sufficiently approximates the real fault centroid, i.e., whether a sufficiently good solution has been found; Expected Information Gain Criterion. This is used to evaluate whether the surrogate model's exploration value for unknown regions has shown diminishing marginal returns, i.e., whether there is still a need to continue searching. The algorithm terminates the iteration through an adaptive triggering mechanism only when any of the criteria are met, thereby avoiding ineffective over-searching.

[0068] For the feature space density criterion This criterion is based on the feature matching fitness calculated in step S3. When the optimal fitness is lower than the preset minimum value, it indicates that the spatiotemporal features of the virtual fault point generation and the real measurement data have formed a highly dense cluster in the feature space. At this point, the positioning error is considered to be less than the system tolerance, and the feature space density criterion is applied. The expression is:

[0069]

[0070] in, As of the end of the In the next iteration, the minimum feature matching fitness recorded in the historical observation set, This is the density threshold.

[0071] Expected information gain criterion This criterion is based on the expected improvement (EI) calculated by the acquisition function in step S4. As the fusion optimization algorithm gradually covers the search space, the EI value will show an oscillating decay trend. When the EI value of several consecutive iterations is lower than a small threshold, it indicates that the model's exploration gains in the remaining unknown regions are insufficient to compensate for the computational cost, and the algorithm enters a search stagnation state.

[0072] Furthermore, to prevent misjudgments caused by local maxima, a sliding window mechanism is used for evaluation; specifically, the expected information gain criterion is employed. The expression is:

[0073]

[0074] in, The set sliding window length, In the past The expected improvement calculated in the next iteration. This is the gain threshold. This formula represents the recent... The average expected gain of the next iteration is below the cutoff threshold, so there is no need to continue iterating.

[0075] At the end of each iteration, the system performs parallel computation. and The state when the feature space compactness criterion And expected information gain criterion When any one of them is true, then the following condition is satisfied. and If any of the following conditions are met, the algorithm is considered to have reached convergence, triggering a termination instruction to stop iteration. The optimal variable combination that minimizes the feature matching fitness is then extracted from the historical sample set. Through the Physical parameters are decoded to obtain fault location results and traveling wave velocity.

[0076] To verify the effectiveness of the method of the present invention, a simulation model topology diagram as shown in Figure 2 was built in PSCAD. Specifically, it is a PSCAD simulation model topology diagram of a four-terminal 230kV 1600km active distribution network. In Figure 2, l1=l2=l3=l4=400km, with a total length of 1600km. The simulation set the fault to occur randomly on the four lines. Traveling wave front signal extraction devices were installed at the four active nodes A, B, C, and D, which can detect the fault traveling wave signal in real time. After wavelet transform, the arrival time of the initial traveling wave is recorded.

[0077] According to the results shown in Table 1, the method provided by this invention can accurately identify faulty line sections at different locations in an active power distribution network, achieve precise distance measurement of the fault point, and provide the corresponding wave velocity estimation range. The positioning error is controlled within 50 meters, indicating that the method has high positioning accuracy and small distance measurement deviation.

[0078] Table 1. Location results using the method provided by this invention under different fault types and transition resistance conditions.

[0079]

[0080] In summary, the power grid fault traveling wave localization method based on the fusion optimization algorithm described in the above embodiments has the following beneficial effects:

[0081] (1) This invention avoids the rigid virtual / real time difference matrix comparison mode in traditional methods. It innovatively adopts a matching mechanism between the virtual fault spatiotemporal feature vector and the target cluster centroid. This idea of ​​transforming fault location into feature matching fitness optimization in the feature space not only reduces the computational dimension and avoids the storage and computational overhead caused by multi-dimensional matrix operations, but also enhances the method's tolerance to noisy data. Combined with a fusion optimization algorithm based on a probabilistic surrogate model, it can quickly lock the global optimal solution in a non-convex, multi-peak fault search space. Compared with traditional exhaustive or deterministic search algorithms, it accelerates the convergence speed and reduces the computational cost, significantly improving the real-time performance and applicability of the location process, and is especially suitable for rapid fault diagnosis of large-scale power grids.

[0082] (2) Multi-scale accurate fitting of complex traveling wave transmission characteristics is achieved, enhancing the robustness of positioning. This invention introduces a multi-scale feature coupling kernel function into the fusion optimization algorithm, replacing the single kernel function model. By adaptively coupling local non-smooth components and global isotropic components, it can not only capture the macroscopic attenuation trend of long-distance traveling wave transmission (global features), but also accurately characterize the local small disturbances caused by wave velocity uncertainty and line impedance changes (local features). Combined with the adaptive hybrid acquisition function developed by dynamic balance exploration and development, as well as the dual convergence criteria, it can dynamically adjust the search confidence boundary according to the sparsity of the feature space, effectively avoiding falling into the local optimum trap, and achieving high-precision adaptive positioning in complex power grid topology environments.

[0083] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for locating traveling wave faults in power grids based on a fusion optimization algorithm, characterized in that, include: Step S1: Establish a dynamic virtual fault framework model of the power grid. In the dynamic virtual fault framework model, randomly generate several virtual fault particles containing position coordinates. Use the virtual fault particles as the initial signal source of the virtual fault traveling wave to construct the initial virtual fault particle cluster. Step S2: Use the shortest path algorithm to calculate the topological path length from each virtual fault particle in the virtual fault particle cluster to all measurement points in the power grid. Combined with the preset traveling wave velocity, calculate the theoretical absolute time when the virtual fault traveling wave arrives at each measurement point. Then map the theoretical absolute time into a virtual fault spatiotemporal feature vector in a high-dimensional feature space. Step S3: Obtain the arrival time of the traveling wave recorded at each measurement point when the real fault occurs. Set the arrival time of the traveling wave as the centroid of the target cluster in the feature space. Calculate the generalized distance between the virtual fault spatiotemporal feature vector and the centroid of the target cluster. Use the generalized distance as the feature matching fitness quantified to characterize the virtual fault particles belonging to the real fault category. Step S4: Construct a non-convex optimization problem with the goal of achieving optimal feature matching fitness. Solve the problem using a fusion optimization algorithm to obtain the fault location result and the traveling wave velocity, thus achieving traveling wave localization. This fusion optimization algorithm uses a probabilistic surrogate model to fit the objective function. The core of the model uses a multi-scale feature coupling kernel function. It adaptively weights and couples local non-smooth components with global isotropic components. Through an adaptive hybrid acquisition function, it dynamically balances global exploration and local development. Furthermore, this fusion optimization algorithm introduces a dual convergence criterion based on expected information gain and feature space compactness to dynamically evaluate the optimization state during iteration. In Step S4, the expression of the multi-scale feature coupling kernel function is: in, For multi-scale feature coupling kernel function, As a multi-scale coupling factor, For local non-smooth components, For globally isotropic components, For the Kronecker function, To observe the noise variance, The global squared exponent. For global feature scale parameters, and The first The moment and the The optimal combination of variables at each time step. It is a local squared exponent. For smoothness parameters, For the Gamma function, For local feature scale parameters, For the second type of modified Bessel function; in step S4, the adaptive hybrid acquisition function. The expression is: in, For dynamic balancing weight function, Represents the mathematical expectation. This indicates taking the maximum value. The fitness of the currently known minimum feature matching. and The Gaussian process is in the th... The mean and standard deviation of the predictions from each iteration. This is the confidence interval adjustment coefficient. This is the inflection point for strategy switching. This is the policy switching rate factor.

2. The power grid fault traveling wave localization method based on fusion optimization algorithm according to claim 1, characterized in that, Step S2 specifically includes: calculating the topological path length from each virtual fault particle in the virtual fault particle cluster to all measurement points in the power grid using the shortest path algorithm, expressed as: in, Indicates virtual fault point To the Measurement points The shortest topological path length, This represents the shortest path algorithm. Virtual fault point The location coordinates within the power grid topology; combined with the preset traveling wave velocity. The theoretical absolute time for the arrival of the virtual fault traveling wave at each measurement point is calculated using the following expression: ;in, For the virtual fault traveling wave to reach the first Measurement points The theoretical absolute time; mapping the theoretical absolute time to a virtual fault spatiotemporal feature vector in a high-dimensional feature space, expressed as: in, Represents the spatiotemporal feature vector of a virtual fault. 、 、 、 They represent The 1st, the 2nd, the 3rd The, the Each feature component The total number of measurement points. Indicates transpose; The expression is: in, The temporal centroid of the virtual time series sequence. For the virtual fault traveling wave to reach the first The theoretical absolute time of each measurement point.

3. The power grid fault traveling wave localization method based on fusion optimization algorithm according to claim 2, characterized in that, Step S3 details This includes: defining the set of arrival times of traveling waves recorded at each measurement point when a real fault occurs. for: in, 、 、 They are the 1st, 2nd, and 3rd respectively. The arrival times of the traveling wave are recorded at each measurement point; using feature space mapping rules, the arrival times of the traveling wave are set as the centroids of the target cluster in the feature space. The expression is: in, 、 、 、 They are the 1st, 2nd, and 3rd respectively. The, the One centroid component; The expression is: in, For the first The arrival time of the traveling wave recorded at each measurement point. For the first The arrival time of the traveling wave recorded at each measurement point. The centroid of the real measurement data is used as the temporal centroid; the generalized distance between the spatiotemporal feature vector of the virtual fault and the centroid of the target cluster is calculated, and the generalized distance is used as the feature matching fitness that quantifies the virtual fault particles belonging to the real fault category. The expression is: in, Indicates feature matching fitness. To optimize the combination of variables, ， For the first The confidence weighting coefficient for each measurement point.

4. The power grid fault traveling wave location method based on fusion optimization algorithm according to claim 3, characterized in that, In step S4, the dual convergence criterion based on expected information gain and feature space compactness includes the feature space compactness criterion. And expected information gain criterion ； Feature space compactness criterion The expression is: in, As of the end of the In the next iteration, the minimum feature matching fitness recorded in the historical observation set, Density threshold; Expected information gain criterion The expression is: in, The set sliding window length, In the past The expected improvement calculated in the next iteration. The gain threshold; when the feature space density criterion And expected information gain criterion If any of the above conditions are met, a termination instruction is triggered, the iteration stops, and the optimal combination of variables that minimizes the feature matching fitness is extracted from the historical sample set. Through the Physical parameters are decoded to obtain fault location results and traveling wave velocity.

Images