Power failure fault locating method and system based on power failure feature library and dynamic deduction

Abstract

The application discloses a power failure fault positioning method and system based on a power failure feature library and dynamic deduction, and belongs to the technical field of power failure positioning, which comprises the following steps: analyzing the physical connection of a distribution network to construct a topological constraint space, constructing a power failure feature library based on historical data and establishing an initial probability distribution; converting real-time research and judgment data into dynamic disturbance operators to drive the probability transfer and multi-path search of a failure state; using potential energy boundary constraints and information entropy measurement to drive the convergence of the state to a deterministic failure feature; analyzing the converged feature into a multi-dimensional phase spectrum and matching it with physical node inherent modes to locate a fault node; calculating a prediction deviation and converting it into a weight update gradient, which is fed back to the power failure feature library to reshape the associated logic. The application adopts a high-dimensional space-time deduction and entropy reduction constraint mechanism, can suppress noise disturbance, realizes accurate fault positioning and model self-adaptive evolution, and improves research and judgment robustness.

Description

A method and system for locating power outage faults based on a power outage feature database and dynamic simulation. Technical Field

[0001] This invention relates to the field of power fault location technology, and in particular to a power fault location method and system based on a power outage feature database and dynamic simulation. Background Technology

[0002] The distribution network is a crucial link in the power system connecting users. Its topology is increasingly large and highly heterogeneous, and its operating environment is complex and variable. Rapid and accurate identification of fault types and location are core requirements for ensuring power supply reliability and system resilience, and for achieving grid self-healing and efficient operation and maintenance. Current traditional methods for distribution network fault assessment often rely on manual line inspections or logical judgments based on simple remote signaling data. For example, patent CN119939490A discloses a diagnostic method integrating multi-source data from renewable energy power plant equipment. This method collects and standardizes equipment operating data, extracts dual-frequency domain fault features to establish a feature library, analyzes energy transfer efficiency by combining power transfer fluctuation values, and then locates the fault source through similarity matching of the static feature library, generating a static fault impact link diagram to achieve hierarchical diagnosis and maintenance decisions for equipment faults.

[0003] However, existing fault location technologies still have many obvious limitations. First, they lack a dynamic evolution perspective. The fault impact link diagrams generated by these technologies are essentially static snapshots based on data at a certain moment, failing to capture the state transitions and path fission patterns of distribution network faults over time slices, making it difficult to dynamically predict fault development trends. Second, they lack anti-interference capabilities. They only perform simple data standardization, lacking effective entropy reduction mechanisms and multi-paradigm arbitration capabilities when facing massive high-entropy random noise disturbances in the distribution network, making them prone to misjudgments. Third, they have a single location dimension. These technologies mainly focus on threshold determination of energy transfer efficiency, without considering the deep correlation logic of faults in topological space and time period, making it difficult to achieve accurate convergence from probabilistic uncertainty to physical determinism in complex and ever-changing power grid environments. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide a power outage fault location method and system based on a power outage feature library and dynamic inference. By employing high-dimensional spatiotemporal inference and entropy reduction constraint mechanisms, it can suppress noise disturbances, achieve accurate fault location and adaptive model evolution, and improve the robustness of judgment.

[0005] The above objectives can be achieved through the following approach:

[0006] The power outage fault location method based on a power outage feature database and dynamic simulation includes the following steps:

[0007] The physical connection relationship of the distribution network is analyzed to construct a topological constraint space that restricts the fault propagation path. Noise disturbances in the power outage data are removed and the data is precipitated into a power outage feature library. Based on the power outage feature library, the correlation logic between the time periodic characteristics and spatial distribution characteristics of power outage events is deconstructed, and the correlation logic is quantified into the probability distribution of the topological constraint space to establish the initial probability field environment.

[0008] Real-time capture of multi-source analysis data and transformation into dynamic perturbation operators that stimulate the initial probability field environment; recursively probabilistic deduction of the state transition and path fission of the fault state in the time slice sequence; and constructs a dynamic scenario network spanned by the fault spatiotemporal evolution trajectory and posterior probability density in the topological constraint space.

[0009] The correlation logic of the power outage feature library is solidified into the potential energy constraint boundary of the dynamic scenario network. The fit between the real-time evolution trajectory and the potential energy constraint boundary is measured. The dynamic scenario network is driven to converge to determinism along the minimum action path in phase space, and fault eigenstates emerge.

[0010] The fault eigenstate is analyzed as a holographic phase spectrum and then subjected to global interference matching with the physical intrinsic modes in the topological constraint space. This excites the energy resonance phase-locking effect of nodes in the topological network and visualizes it as the fault spatiotemporal singularity of the energy dissipation extreme value.

[0011] The holographic deviation of the spatiotemporal singularity of the fault relative to the real state of the physical entity of the distribution network is captured, encoded into a cognitive residual vector and transformed into an inverse negative entropy flow. This is then injected into the initial probability field environment and the power outage feature library. The probability distribution of the power outage feature library is updated by reshaping the gradient, enabling an adaptive phase transition in the topological constraint space and the dynamic scenario network.

[0012] Preferably, quantifying the association logic into a probability distribution in the topological constraint space to establish the initial probability field environment includes:

[0013] Using gradient boosting nonlinear decision logic as a negative entropy filter, gain filtering and noise stripping are performed on power outage data, and effective fault information is aggregated and precipitated into a power outage feature library.

[0014] Neighborhood feature aggregation operation is performed on the physical connection relationship of the distribution network to extract node feature vectors to define the geometric boundary of the topological constraint space, and the time-series evolution mode representing the time periodicity is deconstructed from the power outage feature library.

[0015] By exploring the coupling relationship between temporal evolution modes and node feature vectors, frequent interaction itemsets are generated and transformed into state transition tensors and prior potential energy surfaces in the topological constraint space, so as to quantify the probability distribution and establish the initial probability field environment.

[0016] Preferably, constructing a dynamic scenario network within the topological constraint space, spanned by the spatiotemporal evolution trajectory of the fault and the posterior probability density, includes:

[0017] Multi-source analysis data is vectorized in the spatiotemporal dimension to generate random excitation vectors with directionality and intensity, which are then mapped to local excitation sources that induce nonlinear distortion of probability distribution within the topological constraint space, thus forming a dynamic perturbation operator.

[0018] Using local excitation sources as evolution singularities, the state transition tensor in the initial probability field environment is driven by dynamic perturbation operators to perform Markov chain diffusion on a continuous time slice sequence, generating parallel fission paths with different fault evolution probabilities.

[0019] By coupling each parallel fission path and the posterior probability density with a tensor product, a dynamic scenario network with a multidimensional bifurcation structure is constructed, where each node of the dynamic scenario network stores the fault state entropy value of the current time slice.

[0020] Preferably, the parallel fission path includes a spatiotemporal topological migration chain and a path entropy flow density, wherein:

[0021] The spatiotemporal topology migration chain is used to map the node jump trajectory of a local excitation source in the topological constraint space recursively over time slices under the driving force of dynamic perturbation operators.

[0022] Path entropy flow density is used to calculate the cumulative information entropy value along the jump trajectory of a node based on the state transition tensor, representing the existence probability weight in the physical evolution process.

[0023] Preferably, the emergent fault eigenstates include:

[0024] The prior potential energy surface is mapped to a heterogeneous scalar potential field on the dynamic scenario network topology. Low potential energy guiding trenches are constructed in regions that conform to the evolution trend of the prior potential energy surface, and high potential energy blocking barriers are constructed in regions that violate the evolution trend of the prior potential energy surface, thus solidifying them into potential energy constraint boundaries.

[0025] Calculate the relative entropy divergence between the posterior probability distribution of the real-time evolution trajectory and the potential energy constraint boundary, and quantify the relative entropy divergence as information impedance, which measures the degree of information flow obstruction.

[0026] By constructing a nonlinear gain modulation function using information impedance, the energy flow in the dynamic scenario network is selectively amplified and attenuated. Positive feedback gain is applied to the path in the impedance matching state to induce coherent superposition of probability amplitudes, and negative feedback damping is applied to the path in the impedance mismatch state to trigger dissipative attenuation of probability amplitudes, driving the state to spontaneously collapse to the fault eigenstate.

[0027] Preferably, information impedance, which measures relative entropy dispersion as the degree of information flow obstruction, includes:

[0028] By using posterior probability density and prior potential surface for manifold alignment, a probabilistic manifold space is constructed to show the distribution differences between real-time evolution states and historical prior logic.

[0029] Within the probability manifold space, the expected integral of the logarithmic difference between the posterior probability density and the prior potential energy surface is calculated to generate the relative entropy divergence of the current state from the evolutionary pattern.

[0030] The relative entropy divergence is transformed into the topological friction coefficient of evolutionary kinetic energy loss to define the information impedance that hinders the spread of fault states along the current path.

[0031] Preferably, before obtaining the faulty spacetime singularity, the following operations are performed:

[0032] The node feature vectors are used as the basis of the spatial manifold and the path entropy flow density are subjected to tensor compaction operation in the spatiotemporal dimension to generate the evolutionary singularity density spectrum of the clustering intensity of fault energy in different topological dimensions.

[0033] By using the evolutionary singularity density spectrum to non-uniformly reconstruct the geometric curvature of the topologically constrained space, a holographic gravitational lens field is formed that induces the eigenstates of faults to collapse in a directional manner in the physical real space.

[0034] Preferably, the energy resonance phase-locked effect that stimulates nodes in the topological network, and the fault spatiotemporal singularity that is visualized as an extreme value of energy dissipation, includes:

[0035] Orthogonal decomposition of the fault eigenstates in the frequency and spatial domains is performed to extract the holographic phase spectrum of the fault type spectral characteristics and spatial phase distribution, and the inherent impedance response characteristics of each physical node in the topological constraint space are extracted as physical intrinsic modes.

[0036] Using a holographic gravitational lens field as a spatial modulation medium, wavefront reconstruction and phase focusing are performed on the holographic phase spectrum, and standing wave interference patterns are formed under the guidance of the physical intrinsic modes.

[0037] By searching for regions of maximum energy density in the standing wave interferogram, identifying the physical nodes where resonant phase-locking occurs, eliminating the probabilistic superposition states in non-fault regions, and confirming the spatiotemporal singularity of the fault.

[0038] Preferably, the adaptive phase transition that causes the topological constraint space and dynamic scenario network to occur includes:

[0039] The physical entities of the distribution network are projected inversely onto the topological constraint space, and the tensor divergence with the spatiotemporal singularity of the fault on the manifold geometry is calculated to construct the cognitive residual vector of the inconsistency between the cognitive state and the physical facts.

[0040] The cognitive residual vector is mapped to a negative gradient field that drives the probability distribution to evolve to a low potential state, and is transformed into a reverse negative entropy flow that counteracts the increase in internal entropy.

[0041] By using the inverse negative entropy flow to perform differential manifold reconstruction on the power outage feature library, structural plastic deformation of the connection weights in the topological constraint space is induced, thus completing the adaptive phase transition of the dynamic scenario network from metastable to steady state.

[0042] A power outage fault location system based on a power outage feature database and dynamic simulation, used to implement the above method, includes:

[0043] The spatiotemporal topology base and probability field construction module is used to analyze the physical connection relationship of the distribution network to construct a topological constraint space that restricts the fault propagation path, remove noise disturbances in the power outage data and precipitate them into a power outage feature library, deconstruct the correlation logic between the time periodic features and spatial distribution features of power outage events based on the power outage feature library, and quantify the correlation logic into the probability distribution of the topological constraint space to establish the initial probability field environment.

[0044] The dynamic disturbance excitation and scenario network deduction module is used to capture multi-source judgment data in real time and transform it into a dynamic disturbance operator that excites the initial probability field environment. By recursively probabilistically deducing the state transition and path fission of the fault state in the time slice sequence, a dynamic scenario network spanned by the fault spatiotemporal evolution trajectory and posterior probability density is constructed in the topological constraint space.

[0045] The potential energy boundary constraint and eigenstate emergence module is used to solidify the correlation logic of the power outage feature library into the potential energy constraint boundary of the dynamic scenario network, measure the fit between the real-time evolution trajectory and the potential energy constraint boundary, respond to the coherent resonance caused by low impedance causal path and the dissipation damping of high impedance random disturbance, drive the dynamic scenario network to converge deterministically along the minimum action path in phase space, and emerge fault eigenstates.

[0046] The holographic phase resonance and spatiotemporal singularity visualization module is used to resolve the fault eigenstate into a holographic phase spectrum and perform global interference matching with the physical intrinsic modes in the topological constraint space to excite the energy resonance phase-locking effect of nodes in the topological network and visualize the fault spatiotemporal singularity as the extreme value of energy dissipation.

[0047] The cognitive residual feedback and system self-organizing evolution module is used to capture the holographic deviation of the spatiotemporal singularity of the fault relative to the real state of the physical entity of the distribution network, encode it into a cognitive residual vector and transform it into an inverse negative entropy flow, inject it into the initial probability field environment and the power outage feature library, and update the probability distribution of the power outage feature library through gradient reshaping, so that the topological constraint space and dynamic scenario network undergo adaptive phase transition.

[0048] The present invention has the following advantages:

[0049] This invention overcomes the limitations of traditional methods based on static feature matching and threshold determination. By constructing a spatiotemporal causal topological space and a dynamic scenario network, it elevates fault location to a path fission and entropy reduction convergence process based on recursive probability deduction. By applying nonlinear dynamic screening to parallel evolution paths using intrinsic potential energy boundaries, it effectively suppresses high-entropy random disturbances and achieves accurate convergence from uncertain probability distributions to deterministic fault eigenstates in complex and ever-changing distribution network environments, thus improving the robustness of the assessment.

[0050] This invention abandons the linear diagnostic logic based solely on energy transfer efficiency and introduces holographic phase resonance and gravitational lensing field mechanisms. By resolving the fault eigenstate into a holographic phase spectrum and performing global interference matching with the physical intrinsic modes, an energy resonance phase-locked effect is excited at specific nodes. This forces the diffuse probability state to collapse instantaneously into an energy dissipation extreme singularity in physical real space, enhancing the ability to capture weak fault features and improving spatiotemporal positioning accuracy. This effectively solves the problems of missed and false diagnoses in low signal-to-noise ratio scenarios. Attached Figure Description

[0051] Figure 1 is a flowchart illustrating the method of the present invention;

[0052] Figure 2 is a distribution diagram of the information gain contribution of the feature dimension in Embodiment 1 of the present invention;

[0053] Figure 3 is a schematic diagram of the multi-dimensional feature association logic consensus degree screening in Embodiment 1 of the present invention;

[0054] Figure 4 is a schematic diagram of the system of the present invention. Detailed Implementation

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0056] Example 1: As shown in Figure 1, the power outage fault location method based on a power outage feature database and dynamic simulation includes the following steps:

[0057] The physical connection relationship of the distribution network is analyzed to construct a topological constraint space that restricts the fault propagation path. Noise disturbances in the power outage data are removed and the data is precipitated into a power outage feature library. Based on the power outage feature library, the correlation logic between the time periodic characteristics and spatial distribution characteristics of power outage events is deconstructed, and the correlation logic is quantified into the probability distribution of the topological constraint space to establish the initial probability field environment.

[0058] Real-time capture of multi-source analysis data and transformation into dynamic perturbation operators that stimulate the initial probability field environment; recursively probabilistic deduction of the state transition and path fission of the fault state in the time slice sequence; and constructs a dynamic scenario network spanned by the fault spatiotemporal evolution trajectory and posterior probability density in the topological constraint space.

[0059] The correlation logic of the power outage feature library is solidified into the potential energy constraint boundary of the dynamic scenario network. The fit between the real-time evolution trajectory and the potential energy constraint boundary is measured. The dynamic scenario network is driven to converge to determinism along the minimum action path in phase space, and fault eigenstates emerge.

[0060] The fault eigenstate is analyzed as a holographic phase spectrum and then subjected to global interference matching with the physical intrinsic modes in the topological constraint space. This excites the energy resonance phase-locking effect of nodes in the topological network and visualizes it as the fault spatiotemporal singularity of the energy dissipation extreme value.

[0061] The holographic deviation of the spatiotemporal singularity of the fault relative to the real state of the physical entity of the distribution network is captured, encoded into a cognitive residual vector and transformed into an inverse negative entropy flow. This is then injected into the initial probability field environment and the power outage feature library. The probability distribution of the power outage feature library is updated by reshaping the gradient, enabling an adaptive phase transition in the topological constraint space and the dynamic scenario network.

[0062] Quantifying the correlation logic into a probability distribution in the topological constraint space to establish the initial probability field environment includes:

[0063] Using gradient boosting nonlinear decision logic as a negative entropy filter, gain filtering and noise stripping are performed on power outage data, and effective fault information is aggregated and precipitated into a power outage feature library.

[0064] Unified preprocessing is performed on outage data from historical events and real-time acquisition, converting noisy raw measurements, user repair records, and switch and protection action logs into structured feature vectors. A nonlinear decision model with a gradient boosting mechanism is introduced as a negative entropy filter. This model can employ any combination of gradient boosting tree models, ensemble learning models, or tree-based boosting classification models. It iteratively fits the residual information between outage labels and features, calculating the contribution of each feature dimension to the fault discrimination information gain. As shown in Figure 2, current fluctuation amplitude and historical fault frequency have high information gain and are identified as key features and retained, while low-gain features such as temperature are treated as noise and filtered out.

[0065] Based on the information gain ranking, an adjustable gain threshold is set. Features with a contribution below this threshold are considered noise or redundant information and removed from the power outage data. Features with high information gain and closely related samples are marked as valid fault information. After the above processing, these high signal-to-noise ratio samples and their features are organized into a power outage feature library for power outage events, providing a data foundation for subsequent temporal and spatial correlation modeling.

[0066] For example, in the operational history of a city's power distribution network, there are numerous records of power outages caused by wind-blown wire breaks, tree-induced short circuits, and switchgear malfunctions. These records include instantaneous fluctuations in current and voltage, as well as manual descriptions of the fault causes entered by maintenance personnel. When performing gain analysis using a negative entropy filter, it was found that features such as "current fluctuation amplitude 5 minutes before the fault," "fault frequency of this feeder in the past three months," and "wind speed level at the time of the fault" have high information gain, while features such as "temperature on the day" and "maintenance team number" contribute less to fault differentiation. Therefore, the negative entropy filter retains high-contribution features and removes low-contribution feature dimensions, compiling the retained samples and features into a power outage feature library for subsequent modeling.

[0067] Neighborhood feature aggregation operation is performed on the physical connection relationship of the distribution network to extract node feature vectors to define the geometric boundary of the topological constraint space, and the time-series evolution mode representing the time periodicity is deconstructed from the power outage feature library.

[0068] The feeders, branches, switchgear, distribution transformers, and critical load nodes of the distribution network are mapped as node elements in a topological constraint space, and line connections are mapped as edge elements, resulting in a graph structure that reflects electrical connections. For each node in this graph structure, its local physical attributes are collected, such as equipment type, rated current, service life, and historical fault count. Simultaneously, the attributes of its directly connected neighboring nodes and the parameters of the connecting lines, such as line length and impedance parameters, are also collected.

[0069] By combining the attributes of the current node with those of several neighboring nodes, a node feature vector describing the node's physical environment and topological position in the distribution network is obtained. Subsequently, based on the outage occurrence times recorded in the outage feature database, time series decomposition or cluster analysis methods are used to analyze the fault occurrence time series, decomposing time components that characterize long-term trends, seasonal cycles, and short-term fluctuations. These time components are then combined to define a time series evolution mode, used to characterize the periodicity of outage events in the time dimension.

[0070] For example, in a 10 kV feeder with multiple branches, each line segment, sectionalizing switch, and distribution transformer node is mapped to a topological constraint space. The node features include information such as equipment type, geographical location, line length, and historical fault count. For a given distribution transformer node, the neighborhood feature aggregation considers not only the transformer's operating years and load factor but also the annual switching count of its adjacent upstream switching nodes and the number of downstream users, resulting in a multi-dimensional node feature vector.

[0071] Meanwhile, based on years of historical power outage records, the outage times were arranged by calendar date. Analysis revealed that the frequency of faults on this feeder increased significantly during the rainy season from July to September each year and during the peak load period in the afternoon. Based on this, two types of time-series evolution modes were extracted: "seasonal high-incidence mode" and "intraday high-incidence mode" for subsequent modeling.

[0072] By exploring the coupling relationship between temporal evolution modes and node feature vectors, frequent interaction itemsets are generated and transformed into state transition tensors and prior potential energy surfaces in the topological constraint space, so as to quantify the probability distribution and establish the initial probability field environment.

[0073] Each power outage event in the power outage feature database is represented as a "temporal evolution mode-node feature vector combination," and the frequency of these combinations in the entire database is counted. Using association rule mining algorithms, frequent patterns in which a certain type of node feature vector repeatedly co-occurs with power outage events under specific temporal evolution modes are identified, and these patterns are organized into frequent interaction itemsets.

[0074] The patterns in the set of frequent interaction terms are regarded as discrete states in the topological constraint space. A prior potential energy surface is constructed based on the frequency of occurrence of each pattern, so that high-frequency patterns correspond to low potential energy regions and low-frequency patterns correspond to high potential energy regions.

[0075] Simultaneously, the successive occurrence relationships between different modes in the power outage event sequence are statistically analyzed to estimate the confidence of state transition from one mode to another, constructing a state transition tensor describing the transition probabilities between states. The prior potential surface characterizes the static prior importance of each state in the topological constraint space, while the state transition tensor characterizes the dynamic propagation tendency of the fault between different node states in the time dimension. Together, they constitute the quantification result of the probability distribution in the topological constraint space, thereby establishing the initial probability field environment.

[0076] Constructing a dynamic scenario network within a topologically constrained space, spanned by the spatiotemporal evolution trajectory of faults and the posterior probability density, includes:

[0077] Multi-source analysis data is vectorized in the spatiotemporal dimension to generate random excitation vectors with directionality and intensity, which are then mapped to local excitation sources that induce nonlinear distortion of probability distribution within the topological constraint space, thus forming a dynamic perturbation operator.

[0078] The data from multiple sources, including monitoring and measurement systems, switches and protection devices, electricity information collection terminals, user repair platforms, and meteorological service platforms, are formatted in a unified manner. Different types of data, such as current, voltage, power, switch position signals, fault indications, user power outage repair coordinates, wind speed and rainfall levels, are reconstructed into time-stamped record sequences.

[0079] For numerical data, normalization is performed to map physical quantities of different dimensions to a unified numerical range. For categorical data, such as switch status categories, fault category labels, and weather type labels, one-hot encoding or similar discrete feature encoding methods are used to convert them into fixed-length discrete feature vectors. Timestamp information and geographic location information are embedded into these feature vectors, so that each encoded feature vector simultaneously contains the time and location of the disturbance event, the magnitude of the physical quantity change, and the event type.

[0080] Each feature vector encoded in the spatiotemporal dimension is treated as a random excitation vector. Based on the geographical or topological location information it contains, this random excitation vector is mapped to a specific node or set of nodes in the topological constraint space. At this node or set of nodes, the initial probability field environment is locally amplified or attenuated, thus forming a local excitation source. By encapsulating the mapping process from random excitation vectors to local excitation sources, a dynamic perturbation operator is obtained to drive the nonlinear distortion of the probability distribution.

[0081] For example, during a power outage event in a thunderstorm, a rapid fluctuation in the current at the end of a 10kV feeder was detected, and the upstream sectionalizing switch changed from closed to tripped. The user repair platform received a large number of outage calls for a specific transformer area, while the meteorological service platform issued a short-term heavy rainfall and lightning warning for the area. Through spatiotemporal vectorization encoding, information such as the current offset at that moment, the sectionalizing switch tripping status, the number of users reporting repairs, the geographical coordinates of the transformer area, and the thunderstorm level were combined into a random excitation vector. Based on the transformer area's topological location in the distribution network, this random excitation vector was mapped to the corresponding node in the topological constraint space, increasing the probability of local faults at that node and its adjacent nodes, making it a local excitation source driving subsequent simulations.

[0082] Using local excitation sources as evolution singularities, the state transition tensor in the initial probability field environment is driven by dynamic perturbation operators to perform Markov chain diffusion on a continuous time slice sequence, generating parallel fission paths with different fault evolution probabilities.

[0083] Taking the node containing the local excitation source as the evolutionary singularity, the initial fault probability of the evolutionary singularity is injected into the state vector of the corresponding node in the state space. Based on the state transition tensor, the update relationship of the node state in each discrete time slice is defined, and it is agreed that the fault state in each time slice can only propagate from the current node to its adjacent nodes with topological connection according to the transition probability given by the state transition tensor.

[0084] State updates based on the Markov property assumption are repeatedly performed on a continuous sequence of time slices: the node failure probability in each time slice depends only on the node failure probability distribution of the previous time slice and the transition probability coefficients given in the state transition tensor, and does not depend on the historical state of earlier time slices. During this diffusion process, the failure state propagates along multiple possible node sequences, each representing a possible failure evolution path. The set of these node sequences, starting from the evolutionary singularity and traversing different nodes in different time slices, is defined as the set of parallel fission paths, used to describe the multiple possibilities of the spatiotemporal evolution of the failure within the topological constraint space.

[0085] By coupling each parallel fission path and the posterior probability density with a tensor product, a dynamic scenario network with a multidimensional bifurcation structure is constructed, where each node of the dynamic scenario network stores the fault state entropy value of the current time slice.

[0086] A dynamic scenario network with a multidimensional bifurcation structure is constructed by tensor product coupling of parallel fission paths and posterior probability densities. Each parallel fission path is represented as an ordered sequence of multiple nodes and time slice indices, and the failure probability distribution of all nodes in each time slice is used as the posterior probability density corresponding to the time slice index. Through tensor product coupling, the failure probability of each node in the parallel fission path at different time slices is extracted from the posterior probability density and aligned with the node sequence of the path, constructing a multidimensional structure containing three types of information: node identifier, time slice index, and failure probability. The multidimensional structures of all parallel fission paths are then uniformly organized into a graph structure, with each "node-time slice" combination considered as a scenario node in the dynamic scenario network, and adjacent "node-time slice" combinations in the parallel fission path considered as directed edges between scenario nodes, thus forming a multidimensional bifurcation structure with temporal hierarchy and path branching relationships.

[0087] For each scenario node in the dynamic scenario network, the fault state entropy value is calculated based on the fault probability value at the corresponding location of the scenario node. The fault state entropy value is used to characterize the degree of uncertainty of the fault state of the scenario node in the current time slice.

[0088] By storing the fault state entropy value in the attributes of the corresponding scenario node, the dynamic scenario network not only expresses the spatiotemporal evolution trajectory and branch structure of the fault, but also explicitly characterizes the state uncertainty level of each scenario node. This provides a data foundation for subsequent path selection and eigenstate emergence based on information entropy and information impedance. As shown in Figure 3, the concentric rings from the inside out represent the feature importance evaluation results based on different dimensions such as probability distribution, temporal mode, spatial topology and association rules. The outermost radially stacked bar chart shows the comprehensive consensus score of each feature under the multidimensional evaluation system.

[0089] Parallel fission paths include spatiotemporal topological migration chains and path entropy flow density, where:

[0090] The spatiotemporal topology migration chain is used to map the node jump trajectory of a local excitation source in the topological constraint space recursively over time slices under the driving force of dynamic perturbation operators.

[0091] A fault evolution path is selected from the set of parallel fission paths, starting from the node where the local excitation source is located. The node sequence of this fault evolution path on continuous time slices is regarded as an ordered pair sequence of "time slice index - node identifier". This ordered pair sequence is sorted according to the time slice index, so that each element clearly corresponds to a certain time slice and its corresponding node in the topological constraint space. Then, based on the graph structure information of the topological constraint space, the edge relationship between node pairs under adjacent time slice indices is confirmed, and the specific topological edge for each node state transition from the current node to the next node is marked.

[0092] Through this process, the node state changes that were originally implicit in the probabilistic evolution process can be explicitly organized into a node jump trajectory that recursively follows the time axis and migrates step by step along the topological edges. This trajectory is defined as a spatiotemporal topological migration chain. The spatiotemporal topological migration chain carries both temporal sequence information and topological positional relationships, and is used to characterize the propagation path and direction of a specific parallel fission path in the topologically constrained space, providing an accurate spatiotemporal skeleton for subsequent calculation of path entropy flux density.

[0093] Path entropy flow density is used to calculate the cumulative information entropy value along the jump trajectory of a node based on the state transition tensor, representing the existence probability weight in the physical evolution process.

[0094] At each time slice, the fault state entropy value corresponding to the current node in the spatiotemporal topology migration chain is obtained. This fault state entropy value is calculated from the fault probability of the node in the dynamic scenario network and the probability distribution related to the fault state, and is used to characterize the degree of uncertainty of the fault state in the current time slice.

[0095] Then, in the temporal order of the spatiotemporal topology migration chain, each node is traversed sequentially from the starting time slice to the ending time slice, and the change in fault state entropy value between adjacent time slices is regarded as the information entropy increment along the jump trajectory of that node. By using pre-defined entropy flow calculation rules, such as accumulating the fault state entropy value or the change in fault state entropy value at each step according to a uniform weight, the cumulative information entropy value on the entire spatiotemporal topology migration chain can be obtained.

[0096] The distribution of the accumulated information entropy value per unit time or per unit topological length is defined as the path entropy flow density. It is used to represent the strength of uncertainty accumulated at different time periods and different topological locations when the fault propagates along the spatiotemporal topological migration chain, thereby quantitatively characterizing the existence probability weight and evolutionary stability of the parallel fission path in the physical evolution process.

[0097] Emergent fault eigenstates include:

[0098] The prior potential energy surface is mapped to an anisotropic scalar potential field on the dynamic scenario network topology. Low potential energy guiding trenches are constructed in regions that conform to the evolution trend of the prior potential energy surface, and high potential energy blocking barriers are constructed in regions that violate the evolution trend of the prior potential energy surface, thus solidifying them into potential energy constraint boundaries.

[0099] The prior potential energy value corresponding to each discrete state is read from the initial probability field environment. Each discrete state is mapped one-to-one with a scenario node in the dynamic scenario network, so that each scenario node obtains a scalar potential energy value. Then, on the topology of the dynamic scenario network, according to the difference in potential energy values ​​between adjacent scenario nodes, a set of paths with gradually decreasing potential energy values ​​is identified. These sets of paths are marked as low-potential-energy guiding trenches, which are used to represent evolutionary channels that conform to the evolution trend of the prior potential energy surface. At the same time, regions with abrupt increases in potential energy values ​​or significantly higher than the average level of surrounding nodes are identified. These regions and their adjacent edges are marked as high-potential-energy barrier walls.

[0100] By fixing and storing the positions of low-potential guiding trenches and high-potential barrier barriers, a potential energy constraint boundary covering the entire dynamic scenario network is formed, providing a static comparison benchmark for evaluating whether the real-time evolution trajectory deviates from the historical evolution pattern.

[0101] Calculate the relative entropy divergence between the posterior probability distribution of the real-time evolution trajectory and the potential energy constraint boundary, and quantify the relative entropy divergence as information impedance, which measures the degree of information flow obstruction.

[0102] Within each time slice, the real-time failure probability distribution of all scenario nodes in the dynamic scenario network is statistically analyzed to obtain the posterior probability distribution of the corresponding time slice. Then, the prior potential energy value recorded in the potential energy constraint boundary is converted into a prior probability distribution, so that each scenario node has both prior and posterior probabilities.

[0103] Next, based on the difference between the posterior probability distribution and the prior probability distribution, the deviation of the real-time evolution state from the historical evolution pattern under each time slice is calculated according to the weighted average principle of logarithmic difference. This deviation is then normalized and aggregated across all scenario nodes in the network to obtain the relative entropy divergence index corresponding to that time slice.

[0104] Using monotonically increasing activation functions commonly used in deep learning, such as the Sigmoid function, the relative entropy divergence index is mapped to an information impedance value. The information impedance value is used to quantitatively represent the degree of obstruction of information flow in the current evolutionary state. The greater the relative entropy divergence, the higher the information impedance.

[0105] By constructing a nonlinear gain modulation function using information impedance, the energy flow in the dynamic scenario network is selectively amplified and attenuated. Positive feedback gain is applied to the path in the impedance matching state to induce coherent superposition of probability amplitudes, and negative feedback damping is applied to the path in the impedance mismatch state to trigger dissipative attenuation of probability amplitudes, driving the state to spontaneously collapse to the fault eigenstate.

[0106] Based on the range of information impedance values, the parallel fission paths in the dynamic scenario network are divided into a set of impedance-matched state paths and a set of impedance-mismatched state paths. The set of impedance-matched state paths corresponds to paths with low information impedance and a high degree of consistency with the evolution trend of the prior potential energy surface, while the set of impedance-mismatched state paths corresponds to paths with high information impedance and a significant deviation from the evolution trend of the prior potential energy surface.

[0107] A nonlinear gain modulation function with information impedance as the independent variable is constructed. A positive feedback gain coefficient greater than one is assigned to the path with lower information impedance, and a negative feedback damping coefficient less than one is assigned to the path with higher information impedance. The probability amplitude of each path is amplified or attenuated iteratively on a continuous time slice.

[0108] As the number of iterations increases, the probability amplitude on the impedance-matched state path set is continuously enhanced through coherent superposition, while the probability amplitude on the impedance-mismatched state path set gradually decreases through dissipative decay. Ultimately, only a very few paths with the lowest total information impedance maintain significant probability amplitudes. The convergence states of these remaining paths in the dynamic scenario network are defined as fault eigenstates. Under these fault eigenstates, the corresponding topological locations and time slice indices are extracted, realizing a spontaneous collapse process from a multi-branch evolution state to a single fault eigenstate.

[0109] Information impedance, which measures relative entropy dispersion as the degree of obstruction to information flow, includes:

[0110] By using posterior probability density and prior potential surface for manifold alignment, a probabilistic manifold space is constructed to show the distribution differences between real-time evolution states and historical prior logic.

[0111] In the dynamic scenario network, each scenario node records both the posterior probability density obtained from real-time inference and the prior probability value mapped from the prior potential energy surface. These two types of probability values ​​are used as the coordinate components of the scenario node in the probability space. Then, using the topological edge relationships of the dynamic scenario network as the adjacency structure, all scenario nodes are embedded into a high-dimensional point set with adjacency relationships, thereby constructing a probabilistic manifold space that simultaneously carries the real-time evolution state and historical prior logic.

[0112] In this probabilistic manifold space, each scenario node retains its positional relationship in the topological constraint space and carries a pair of prior and posterior probabilities, providing a unified geometric carrier for subsequent measurement of the difference in their distributions.

[0113] Within the probability manifold space, the expected integral of the logarithmic difference between the posterior probability density and the prior potential energy surface is calculated to generate the relative entropy divergence of the current state from the evolutionary pattern.

[0114] For each scenario node, the logarithmic difference between its posterior probability density and its corresponding prior probability value is calculated, and this logarithmic difference is used as the local distribution deviation index for that scenario node. Subsequently, based on the importance of the scenario node in the dynamic scenario network, such as the magnitude of the failure probability or its frequency of occurrence in parallel fission paths, weights are assigned to each scenario node. The local distribution deviation indices of all scenario nodes are then weighted, summed, and normalized to obtain the relative entropy divergence, which characterizes the degree of deviation of the current evolutionary state of the entire network from the overall prior evolutionary pattern. The larger the relative entropy divergence value, the more significant the distributional difference between the current real-time evolutionary state and historical statistical patterns.

[0115] The relative entropy divergence is transformed into the topological friction coefficient of evolutionary kinetic energy loss to define the information impedance that hinders the spread of fault states along the current path.

[0116] The relative entropy divergence is mapped to a dimensionless topological friction coefficient, such that a larger relative entropy divergence corresponds to a higher topological friction coefficient. The topological friction coefficient is interpreted as the proportion of effective kinetic energy loss per unit step when the fault state propagates along the current evolution path: the larger the topological friction coefficient, the more "evolutionary energy" is required to advance along the path to counteract the resistance caused by deviation from the prior laws.

[0117] Based on this topological friction coefficient, an information impedance is defined for each parallel fission path. The information impedance is used as a quantitative indicator to measure the ease with which information flow propagates on the path. In the subsequent nonlinear gain modulation process, stronger probability attenuation is applied to the path with larger information impedance, and stronger probability amplification is applied to the path with smaller information impedance.

[0118] Before obtaining the faulty spacetime singularity, perform the following operations:

[0119] The node feature vectors are used as the basis of the spatial manifold and the path entropy flow density are subjected to tensor compaction operation in the spatiotemporal dimension to generate the evolutionary singularity density spectrum of the clustering intensity of fault energy in different topological dimensions.

[0120] The node feature vectors are considered as the spatial manifold basis describing the topological constraint space. Each node feature vector corresponds to a base point in the topological constraint space, and multiple topological dimension features such as equipment type, service life, line length, load characteristics, and historical failure count are recorded at this base point. The calculated path entropy flow density is distributed to each node and time slice according to the correspondence between parallel fission paths and node sequences, so that each "node-time slice" combination carries the uncertainty information accumulated along the jump trajectory of that node.

[0121] The spatial manifold basis of node feature vectors is multidimensionally inner-producted or weighted summed with path entropy flux density over the time dimension. The path entropy flux density at different time slices is then projected onto each feature component of the corresponding node feature vector to obtain the fault energy accumulation intensity at each topological feature dimension. This process is repeated for all nodes and all time slices to form an energy distribution sequence indexed by "topological feature dimension". This energy distribution sequence is defined as the evolutionary singularity density spectrum, used to characterize the accumulation preference and strength differences of fault energy at different topological feature dimensions during the current fault evolution process.

[0122] For example, in a 10 kV distribution line containing multiple branches and several important loads, the node feature vector includes topological dimension features such as "equipment type," "operating years," "number of branch lines," and "number of historical faults." After calculating the path entropy flow density of multiple parallel fission paths using path entropy flow density, in the spatiotemporal tensor compaction operation, the path entropy flow density of each parallel fission path in each time slice is accumulated along the node sequence and projected onto the "operating years" and "number of historical faults" dimensions of the corresponding node feature vector.

[0123] After superimposing multiple parallel fission paths, it was found that the energy value of the evolutionary singularity density spectrum was significantly higher than that of other feature dimensions in the node feature dimension of "long service life and many historical failures". This indicates that in the current failure evolution process, the failure energy is more concentrated near old and frequently failing equipment, thus providing a significant feature direction for subsequent spatial geometric reconstruction.

[0124] By using the evolutionary singularity density spectrum to non-uniformly reconstruct the geometric curvature of the topologically constrained space, a holographic gravitational lens field is formed that induces the eigenstates of faults to collapse in a directional manner in the physical real space.

[0125] Based on the evolutionary singularity density spectrum, the energy accumulation intensity value corresponding to each node is extracted. The topologically constrained space is a weighted graph consisting of a set of nodes V and a set of edges E, where the initial connection impedance between any two nodes i and j is denoted as... Perform non-uniform weighted reconstruction based on feature density to evolve the normalized energy density values ​​of each node in the singularity density spectrum. and Using the following nonlinear impedance modulation formula as input, the reconstructed equivalent connection impedance is calculated. :

[0126] ;

[0127] in, The equivalent connection impedance between node i and node j after reconstruction; The initial connection impedance is determined by the physical line length or electrical impedance. Let be the energy density values ​​of node i and node j in the evolutionary singularity density spectrum; This is the focusing gain coefficient, used to adjust the sensitivity to high-density regions; It is a non-linear exponent used to control the steepness of the gradient for weight adjustment.

[0128] Calculated using this formula, for node regions with higher energy density, the denominator is larger, resulting in a higher equivalent interconnect impedance. The smaller the value, the more likely it is to become a low-impedance trap in the mathematical structure of the topology.

[0129] This will include the reconstructed equivalent connection impedance of all node pairs. The weighted adjacency matrix is ​​defined as the holographic gravitational lensing field. In dynamic simulation, the probability flow of the fault eigenstates will naturally calculate the convergence of low-impedance regions based on the shortest path, thereby achieving spatial focusing and fine resolution in a physical sense.

[0130] For example, in a 10 kV distribution line scenario, when the evolution singularity density spectrum shows that the region with the combination of node characteristics such as "long service life", "many historical faults" and "many branch lines" has a high energy density, the effective path length between nodes with these characteristics is shortened through non-uniform geometric reconstruction, making it easier to "sink" to these high energy density regions from any ordinary node along the probability propagation path.

[0131] Simultaneously, the effective path length is appropriately lengthened for node combinations with shorter operating years, fewer historical faults, and fewer branch lines, thus shortening the residence time of the probability flow in these regions. In subsequent dynamic simulations, the probability amplitude corresponding to the fault eigenstate propagates in the holographic gravitational lensing field, producing a significant focusing effect near the aforementioned high-energy-density regions. Ultimately, it collapses into a few physical nodes that satisfy the evolutionary singularity density spectrum characteristics, thereby achieving high spatial resolution fault location results even with multiple candidate segments.

[0132] The energy resonance phase-locking effect that stimulates nodes in a topological network, visualized as a fault-space singularity with extreme energy dissipation, includes:

[0133] Orthogonal decomposition of the fault eigenstates in the frequency and spatial domains is performed to extract the holographic phase spectrum of the fault type spectral characteristics and spatial phase distribution, and the inherent impedance response characteristics of each physical node in the topological constraint space are extracted as physical intrinsic modes.

[0134] The fault eigenstates are discretized in both the time and topological dimensions to obtain the fault eigenstate time series at each physical node. For each physical node, multidimensional Fourier transform, wavelet transform, or other time-frequency analysis methods commonly used in this field are employed to decompose the fault eigenstate time series into several frequency components and their corresponding phase distributions, resulting in a complex-valued spectral function in the two-dimensional frequency nodes.

[0135] The portion of the complex-valued spectral function that simultaneously carries frequency amplitude information and spatial phase information is defined as the holographic phase spectrum, which is used to characterize the spectral characteristics of the fault type and the relative phase relationship of the fault disturbance at each node in the topological constraint space.

[0136] Meanwhile, based on the parameters of the primary equipment of the distribution network, the electrical model of each physical node in the topological constraint space is established, and the response curves of impedance amplitude and phase with frequency are organized into physical intrinsic modes, which are used to characterize the inherent resonance characteristics of each physical node to fault disturbances at a specific frequency.

[0137] Using a holographic gravitational lens field as a spatial modulation medium, wavefront reconstruction and phase focusing are performed on the holographic phase spectrum, and standing wave interference patterns are formed under the guidance of the physical intrinsic modes.

[0138] Holographic phase spectrum is spatially modulated using a holographic gravitational lensing field. The curvature or weight of the holographic gravitational lensing field at each physical node is used as a modulation factor, which is applied to the frequency components of the holographic phase spectrum corresponding to that node. This differentially delays and deflects the phase at different nodes, thereby completing wavefront reconstruction and phase focusing. The reconstructed holographic phase spectrum is represented on the frequency node plane as a set of complex-valued spectral components that are spatially guided, amplified, or suppressed.

[0139] At each physical node, the holographic phase spectrum modulated by the holographic gravitational lens field is superimposed with the physical intrinsic mode of that node in the frequency domain. The interference intensity of that node in the full frequency band is calculated by integral form, yielding the interference intensity function:

[0140] ;

[0141] in, The interference intensity of physical node n represents the degree of resonant accumulation of fault energy at that node; n represents the identifier of the physical node in the topological constraint space. Represents the angular frequency variable; This indicates that after being modulated by a holographic gravitational lens field, at a frequency The complex-valued component of the holographic phase spectrum corresponding to physical node n has an amplitude that reflects the energy intensity of the fault signal at that frequency and a phase that reflects the phase shift of the fault disturbance at that node. This represents the physical intrinsic mode of physical node n at frequency. The complex impedance response component under the given conditions has an amplitude that reflects the node's sensitivity to the frequency disturbance, and a phase that reflects the node's phase response to the frequency disturbance.

[0142] The physical meaning of the interference intensity function lies in measuring the degree of coherent superposition of the "holographic wave carrying fault information" and the "nodal intrinsic response wave" across the entire frequency band. When both satisfy amplitude matching and phase matching at multiple key frequencies simultaneously, the integral result increases significantly, indicating that the node is more prone to resonant phase-locking of fault energy. In practical implementation, the frequency range can be discretized into several frequency sampling points, and the integration can be approximated with a frequency step size. This discretization is implemented as a weighted summation over a finite number of frequency points to meet the requirements of engineering computational efficiency.

[0143] By searching for regions of maximum energy density in the standing wave interferogram, identifying the physical nodes where resonant phase-locking occurs, eliminating the probabilistic superposition states in non-fault regions, and confirming the spatiotemporal singularity of the fault.

[0144] Based on the interference intensity function, the interference intensity of each physical node in the full frequency band is mapped to the energy density index of that node, and an energy density distribution map is formed on the geometric structure of the topological constraint space. This energy density distribution map is the standing wave interferogram.

[0145] An extremum search process is performed on the standing wave interferogram to identify local maxima regions where the energy density is significantly higher than the average level of the surrounding nodes. The variation trend of the interference intensity gradient between adjacent physical nodes is analyzed to eliminate isolated spikes caused by random noise and retain energy density maxima regions with stable gradient structures.

[0146] By combining the fault probability and fault state entropy value of the corresponding node in the dynamic scenario network, cross-screening is performed on these energy density maxima regions, and physical nodes that simultaneously satisfy high interference strength, high prior probability, and low fault state entropy value are marked as physical nodes that have undergone resonance phase-locking.

[0147] For physical nodes that are not selected, their corresponding probability superposition states are regarded as background states of non-fault regions and are eliminated or significantly suppressed in subsequent positioning outputs. This way, a few physical nodes that meet the resonance phase-locking conditions are retained as energy dissipation extreme points. The spatial coordinates of these physical nodes and the time slice indexes corresponding to the fault eigenstates are jointly defined as fault spatiotemporal singularities, realizing a concrete mapping from the high-dimensional probability domain to the physical real space.

[0148] For example, in a 10 kV distribution line containing multiple sectionalizing switches and multiple distribution transformers, a set of candidate fault sections is obtained after dynamic simulation and fault eigenstate emergence process. After orthogonal decomposition of the fault eigenstates in the frequency and spatial domains, the holographic phase spectrum exhibits obvious phase clustering characteristics near a certain transformer area in the frequency band of 150 Hz to 250 Hz; at the same time, the physical intrinsic mode calculation results show that the cable joint nodes connected to this transformer area exhibit low impedance amplitude and specific phase response in the above frequency band.

[0149] After applying a holographic gravitational lens field to the holographic phase spectrum, the phase of the aforementioned frequency band is focused. At the cable joint node in this distribution area, the modulated holographic phase spectrum components resonate highly with the physical intrinsic modes. By discretizing and summing the interference intensity within the frequency range, the interference intensity function of this node is found to be significantly higher than that of other nodes on the same branch.

[0150] Mapping the interference intensity of all nodes to a standing wave interferogram clearly reveals a region of significant energy density maxima at the cable joint node. Combining the posterior fault probability and low fault state entropy value provided by the dynamic scenario network, the spatial location of the cable joint node and the time slice of the fault eigenstate are output as a pair of spatiotemporal coordinates, confirming it as the spatiotemporal singularity of the fault in this power outage event, providing clear location results for on-site maintenance personnel.

[0151] Adaptive phase transitions in topologically constrained spaces and dynamic scenario networks include:

[0152] The physical entities of the distribution network are projected inversely onto the topological constraint space, and the tensor divergence with the spatiotemporal singularity of the fault on the manifold geometry is calculated to construct the cognitive residual vector of the inconsistency between the cognitive state and the physical facts.

[0153] After the fault handling is completed, the actual fault location, actual fault type, and time information of fault occurrence and resolution are obtained from the dispatch records and on-site emergency repair records after manual verification, and the actual fault status is used as the physical fact benchmark.

[0154] Based on the primary wiring diagram of the distribution network and geographical information, the physical fact benchmark is mapped to the corresponding physical node and its adjacent node set in the topological constraint space to obtain the benchmark state point representing the actual fault state.

[0155] Simultaneously, the position of the visualized spatiotemporal singularity of the fault in the topological constraint space is read from the fault eigenstate derivation results, and this position is regarded as the cognitive state point. Then, on this manifold of the topological constraint space, the geometric deviation of the cognitive state point from the physical fact benchmark state point is calculated by comprehensively considering the electrical connection relationship between nodes, path length, and local curvature introduced by the previous geometric reconstruction. This geometric deviation is then decomposed into multiple topological feature dimensions to form a multidimensional vector with direction and magnitude.

[0156] This multidimensional vector is the cognitive residual vector, which is used to quantitatively characterize the inconsistency between the spatiotemporal singularity of the fault output by the current dynamic scenario network and the actual fault state.

[0157] The cognitive residual vector is mapped to a negative gradient field that drives the probability distribution to evolve to a low potential state, and is transformed into a reverse negative entropy flow that counteracts the increase in internal entropy.

[0158] The calculated cognitive residual vector is denoted as... Each component represents the deviation of the current dynamic scenario network from the physical fact benchmark in a specific feature dimension. A gradient-based parameter correction model is constructed to establish the negative gradient field. The cognitive potential function, i.e., the loss function, is defined. ,in The set of parameters to be corrected, including feature weights and state transition probabilities, is defined in the power outage feature library and topological constraint space. The parameter update increment is calculated using the following inverse negative entropy flow generation formula:

[0159] ;

[0160] And the parameter update formula:

[0161] ;

[0162] in, Defined as the inverse negative entropy flow, i.e., the parameter-corrected gradient vector; For adaptive learning step size; This represents the gradient of the potential energy function with respect to the parameters; Let J be the cognitive residual vector; J is the Jacobian matrix, representing the sensitivity of parameter changes to the output residual; T represents the matrix transpose operation. , Before and after the update By mapping the cognitive residual vector to a vector field flowing in the parameter space, this vector field... The direction strictly points to the potential energy function The direction of the fastest descent, i.e. the negative gradient direction, has the physical significance of injecting structured error correction information that is opposite to the current deviation, in order to offset the "entropy increase" caused by the internal model deviation, thereby driving the probability distribution of the power outage feature library to approach the real physical law.

[0163] By using the inverse negative entropy flow to perform differential manifold reconstruction on the power outage feature library, structural plastic deformation of the connection weights in the topological constraint space is induced, thus completing the adaptive phase transition of the dynamic scenario network from metastable to steady state.

[0164] Using reverse negative entropy flow as the update driving force, the weights of features in the power outage feature library that are highly correlated with the actual fault mode are increased, while the weights of features that contribute significantly to cognitive bias in the simulation but do not appear or appear with significantly low frequency in the actual fault are reduced. This results in a differential-level weight rearrangement in the feature space of the power outage feature library, achieving a local reshaping of the feature statistical distribution.

[0165] The reverse negative entropy flow is projected onto each node and its edge weights in the topological constraint space. The state transition probability of the edge that highly overlaps with the actual fault path is appropriately increased, and the state transition probability of the edge that highly overlaps with the erroneous parallel fission path is appropriately decreased. At the same time, some low-contribution edges are weakened when necessary, so that the state transition tensor of the entire topological constraint space undergoes continuous and smooth parameter updates along the direction of the reverse negative entropy flow.

[0166] As feedback from multiple fault events accumulates, the aforementioned differential manifold reconstruction is gradually superimposed within the power outage feature library and topological constraint space, triggering plastic deformation of connection weights and state transition probabilities at the structural level. This causes the dynamic scenario network to evolve from a metastable state containing historical cognitive biases to a steady state that is more consistent with physical facts and statistical laws, thereby completing the adaptive phase transition of the dynamic scenario network. This is reflected in higher convergence speed and higher positioning accuracy in subsequent fault location tasks.

[0167] Example 2: A power outage fault location system based on a power outage feature database and dynamic simulation, used to implement the method in Example 1, as shown in Figure 4. The system includes:

[0168] The spatiotemporal topology base and probability field construction module is used to analyze the physical connection relationship of the distribution network to construct a topological constraint space that restricts the fault propagation path, remove noise disturbances in the power outage data and precipitate them into a power outage feature library, deconstruct the correlation logic between the time periodic features and spatial distribution features of power outage events based on the power outage feature library, and quantify the correlation logic into the probability distribution of the topological constraint space to establish the initial probability field environment.

[0169] The dynamic disturbance excitation and scenario network deduction module is used to capture multi-source judgment data in real time and transform it into a dynamic disturbance operator that excites the initial probability field environment. By recursively probabilistically deducing the state transition and path fission of the fault state in the time slice sequence, a dynamic scenario network spanned by the fault spatiotemporal evolution trajectory and posterior probability density is constructed in the topological constraint space.

[0170] The potential energy boundary constraint and eigenstate emergence module is used to solidify the correlation logic of the power outage feature library into the potential energy constraint boundary of the dynamic scenario network, measure the fit between the real-time evolution trajectory and the potential energy constraint boundary, respond to the coherent resonance caused by low impedance causal path and the dissipation damping of high impedance random disturbance, drive the dynamic scenario network to converge deterministically along the minimum action path in phase space, and emerge fault eigenstates.

[0171] The holographic phase resonance and spatiotemporal singularity visualization module is used to resolve the fault eigenstate into a holographic phase spectrum and perform global interference matching with the physical intrinsic modes in the topological constraint space to excite the energy resonance phase-locking effect of nodes in the topological network and visualize the fault spatiotemporal singularity as the extreme value of energy dissipation.

[0172] The cognitive residual feedback and system self-organizing evolution module is used to capture the holographic deviation of the spatiotemporal singularity of the fault relative to the real state of the physical entity of the distribution network, encode it into a cognitive residual vector and transform it into an inverse negative entropy flow, inject it into the initial probability field environment and the power outage feature library, and update the probability distribution of the power outage feature library through gradient reshaping, so that the topological constraint space and dynamic scenario network undergo adaptive phase transition.

[0173] Example 3: A power outage fault location device based on a power outage feature library and dynamic simulation, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. The method in Example 1 is implemented by executing the program through the processor.

Claims

1. A power outage fault location method based on a power outage feature database and dynamic simulation, characterized in that, The steps include: This study analyzes the physical connections of the distribution network to construct a topological constraint space that restricts fault propagation paths. Noise disturbances in outage data are removed and precipitated into an outage feature library. Based on this feature library, the correlation logic between the temporal periodicity and spatial distribution characteristics of outage events is deconstructed, and this correlation logic is quantified into a probability distribution within the topological constraint space to establish the initial probability field environment. Multi-source assessment data is captured in real-time and transformed into dynamic disturbance operators that stimulate the initial probability field environment. Through recursive probability deduction, the state transitions and path fissions of fault states in time-slice sequences are deduced. Within the topological constraint space, a dynamic scenario network spanned by the spatiotemporal evolution trajectory of the fault and the posterior probability density is constructed. The correlation logic of the outage feature library is solidified into the potential energy constraint boundary of the dynamic scenario network, and the real-time evolution trajectory and potential are measured. The system can constrain the fit of the boundary, respond to the coherent resonance caused by low-impedance causal paths and the dissipation damping of high-impedance random disturbances, drive the dynamic scenario network to converge deterministically along the minimum action path in phase space, and generate fault eigenstates. The fault eigenstates are analyzed as holographic phase spectra and subjected to global interference matching with the physical intrinsic modes in the topological constraint space, which excites the energy resonance phase-locking effect of nodes in the topological network and visualizes them as fault spatiotemporal singularities with extreme energy dissipation values. The system captures the holographic deviation of the fault spatiotemporal singularities relative to the real state of the distribution network physical entity, encodes it as a cognitive residual vector and transforms it into an inverse negative entropy flow, injects it into the initial probability field environment and the outage feature library, and updates the probability distribution of the outage feature library by gradient reshaping, so that the topological constraint space and the dynamic scenario network undergo adaptive phase transition.

2. The power outage fault location method based on a power outage feature database and dynamic deduction as described in claim 1, characterized in that, Quantifying the correlation logic into a probability distribution in the topological constraint space to establish the initial probability field environment includes: using gradient boosting nonlinear decision logic as a negative entropy filter to perform gain filtering and noise stripping on power outage data, aggregating and precipitating effective fault information into a power outage feature library; performing neighborhood feature aggregation operations on the physical connection relationships of the distribution network, extracting node feature vectors to define the geometric boundary of the topological constraint space, and deconstructing the time-series evolution modes representing time periodicity from the power outage feature library; mining the coupling correlation between the time-series evolution modes and node feature vectors, generating frequent interaction itemsets and transforming them into state transition tensors and prior potential energy surfaces in the topological constraint space to complete the quantization of the probability distribution and establish the initial probability field environment.

3. The power outage fault location method based on a power outage feature database and dynamic deduction as described in claim 2, characterized in that, Constructing a dynamic scenario network within a topologically constrained space, spanned by the spatiotemporal evolution trajectory of a fault and the posterior probability density, involves: vectorizing multi-source assessment data in a spatiotemporal dimension to generate random excitation vectors with directionality and intensity, which are then mapped to local excitation sources within the topologically constrained space that induce nonlinear distortions in the probability distribution, thus forming a dynamic perturbation operator; using the local excitation sources as evolution singularities, the dynamic perturbation operator drives the state transition tensor in the initial probability field environment to perform Markov chain diffusion on a continuous time slice sequence, generating parallel fission paths with different fault evolution probabilities; and coupling each parallel fission path and the posterior probability density with a tensor product to construct a multidimensional bifurcation structure dynamic scenario network, where each node of the dynamic scenario network stores the fault state entropy value of the current time slice.

4. The power outage fault location method based on a power outage feature database and dynamic deduction according to claim 3, characterized in that, The parallel fission path includes a spatiotemporal topological migration chain and a path entropy flow density. The spatiotemporal topological migration chain is used to map the node jump trajectory of a local excitation source recursively over time slices in the topologically constrained space under the drive of a dynamic perturbation operator. The path entropy flow density is used to calculate the cumulative information entropy value along the node jump trajectory based on the state transition tensor, representing the existence probability weight in the physical evolution process.

5. The power outage fault location method based on a power outage feature database and dynamic deduction according to claim 2, characterized in that, Emergent fault eigenstates include: mapping the prior potential energy surface to an heterogeneous scalar potential field on the dynamic scenario network topology; constructing low-potential-energy guiding trenches in regions conforming to the evolution trend of the prior potential energy surface; constructing high-potential-energy blocking barriers in regions violating the evolution trend of the prior potential energy surface; solidifying these barriers into potential energy constraint boundaries; calculating the relative entropy divergence between the posterior probability distribution of the real-time evolution trajectory and the potential energy constraint boundary; quantifying the relative entropy divergence into information impedance, which measures the degree of information flow obstruction; constructing a nonlinear gain modulation function using the information impedance; selectively amplifying and attenuating the energy flow in the dynamic scenario network; applying positive feedback gain to paths in impedance-matched states to induce coherent superposition of probability amplitudes; applying negative feedback damping to paths in impedance-mismatched states to trigger dissipative attenuation of probability amplitudes; and driving the state to spontaneously collapse to the fault eigenstate.

6. The power outage fault location method based on a power outage feature database and dynamic deduction according to claim 5, characterized in that, The information impedance that measures relative entropy divergence as the degree of information flow obstruction includes: using posterior probability density and prior potential energy surface for manifold alignment to construct a probability manifold space of the distribution difference between the real-time evolution state and the historical prior logic; within the probability manifold space, calculating the expected integral of the logarithmic difference between the posterior probability density and the prior potential energy surface to generate the relative entropy divergence of the current state from the evolution law; and converting the relative entropy divergence into the topological friction coefficient of evolutionary kinetic energy loss to define the information impedance that hinders the spread of the fault state along the current path.

7. The power outage fault location method based on a power outage feature database and dynamic deduction according to claim 4, characterized in that, Before obtaining the spatiotemporal singularity of the fault, perform the following operations: use the node feature vector as the basis of the spatial manifold and perform tensor compaction operation on the spatiotemporal dimension with the path entropy flux density to generate the evolutionary singularity density spectrum of the intensity of fault energy accumulation in different topological dimensions; use the evolutionary singularity density spectrum to non-uniformly reconstruct the geometric curvature of the topological constraint space to form a holographic gravitational lens field that induces the fault eigenstate to undergo directional collapse in the physical real space.

8. The power outage fault location method based on a power outage feature database and dynamic deduction according to claim 7, characterized in that, The process of stimulating the energy resonance phase-locking effect of nodes in a topological network, and visualizing it as a fault spatiotemporal singularity with extreme energy dissipation, involves: performing orthogonal decomposition of the fault eigenstates in the frequency and spatial domains; extracting the holographic phase spectrum of the fault type spectral characteristics and spatial phase distribution; and extracting the inherent impedance response characteristics of each physical node in the topological constraint space as the physical intrinsic modes; using a holographic gravitational lens field as a spatial modulation medium to perform wavefront reconstruction and phase focusing on the holographic phase spectrum; and forming a standing wave interferogram under the guidance of the physical intrinsic modes; searching for the energy density maxima region in the standing wave interferogram; identifying the physical nodes where resonance phase-locking occurs; eliminating the probabilistic superposition states in non-fault regions; and confirming the fault spatiotemporal singularity.

9. The power outage fault location method based on a power outage feature database and dynamic deduction according to claim 1, characterized in that, The adaptive phase transition of the topological constraint space and the dynamic scenario network includes: projecting the physical entities of the distribution network inversely onto the topological constraint space and calculating the tensor divergence of the fault spatiotemporal singularity on the manifold geometry to construct a cognitive residual vector that reflects the inconsistency between the cognitive state and the physical facts; mapping the cognitive residual vector to a negative gradient field that drives the probability distribution to evolve towards a low potential state and transforming it into a reverse negative entropy flow that offsets the increase in internal entropy; and using the reverse negative entropy flow to perform differential manifold reconstruction on the outage feature library, inducing structural plastic deformation of the connection weights within the topological constraint space, thereby completing the adaptive phase transition of the dynamic scenario network from a metastable state to a steady state.

10. A power outage fault location system based on a power outage feature database and dynamic deduction, used to implement the power outage fault location method based on a power outage feature database and dynamic deduction as described in any one of claims 1-9, characterized in that, include: The spatiotemporal topology and probability field construction module is used to analyze the physical connection relationships of the distribution network to construct a topological constraint space that restricts fault propagation paths. It removes noise disturbances from outage data and precipitates them into an outage feature library. Based on this feature library, it deconstructs the correlation logic between the temporal periodic characteristics and spatial distribution characteristics of outage events, and quantifies this correlation logic into a probability distribution of the topological constraint space to establish the initial probability field environment. The dynamic disturbance excitation and scenario network deduction module is used to capture multi-source assessment data in real time and transform it into dynamic disturbance operators that excite the initial probability field environment. Through recursive probability deduction of the state transitions and path fission of fault states in time-slice sequences, it constructs a dynamic scenario network spanned by the spatiotemporal evolution trajectory of the fault and the posterior probability density within the topological constraint space. The potential energy boundary constraint and eigenstate emergence module solidifies the correlation logic of the outage feature library into the potential energy constraint boundary of the dynamic scenario network, measuring the actual... The degree of fit between the temporal evolution trajectory and the potential energy constraint boundary, responding to the coherent resonance caused by the low-impedance causal path and the dissipation damping of the high-impedance random disturbance, drives the dynamic scenario network to converge deterministically along the minimum action path in the phase space, giving rise to fault eigenstates; the holographic phase resonance and spatiotemporal singularity visualization module is used to resolve the fault eigenstate into a holographic phase spectrum and perform global interference matching with the physical intrinsic modes in the topological constraint space, stimulating the energy resonance phase-locking effect of nodes in the topological network, and visualizing it as the fault spatiotemporal singularity of the energy dissipation extremum; the cognitive residual feedback and system self-organizing evolution module is used to capture the holographic deviation of the fault spatiotemporal singularity relative to the real state of the distribution network physical entity, encode it into a cognitive residual vector and transform it into an inverse negative entropy flow, inject it into the initial probability field environment and the power outage feature library, and update the probability distribution of the power outage feature library through gradient reshaping, so that the topological constraint space and the dynamic scenario network undergo an adaptive phase transition.

Images