A method for power distribution line fault early warning and analysis

Abstract

This application provides a method for early warning and analysis of power distribution line faults, including: calculating the probability of each transformation path based on the type transformation direction sequence, integrating the transformation time interval and path weight allocation; if a path corresponds to a hidden feature marker and its probability is higher than a preset probability threshold, then a high-risk transformation path is identified; through the high-risk transformation path, a fault evolution map is generated; based on the node connection density and node influence radius, graph structure analysis technology is used to sort out the path cycle pattern and determine the fault evolution logic chain; based on the risk decision criteria, the power distribution system monitoring rules are updated; the monitoring threshold is adjusted based on environmental impact factors and signal interference intensity; and the monitoring logic is optimized by combining hidden feature markers to obtain a refined fault identification mechanism.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method for early warning and analysis of power distribution line faults. Background Technology

[0002] The safe and stable operation of the power distribution system is a crucial cornerstone for ensuring power supply. Especially in modern cities and industrial sectors, any failure can lead to serious economic losses and social impacts.

[0003] Studying the identification and evolution patterns of power distribution equipment faults is not only related to the reliability of the power system, but also directly related to public safety and quality of life.

[0004] Therefore, exploring the dynamic changes and potential transformation paths of fault types has become a crucial direction that cannot be ignored in improving the level of power operation and maintenance.

[0005] However, in current power distribution fault management, many methods are often limited to the detection and handling of single faults, ignoring the complex evolutionary relationships that may exist between faults.

[0006] This approach struggles to capture the dynamic process of a fault transforming from one form to another, especially in complex operating environments. It lacks the ability to predict fault development trends, often leaving maintenance personnel in a reactive state and missing the best opportunity for intervention.

[0007] A deeper technical challenge lies in the fact that the conversion of fault types in power distribution systems is affected by a variety of electrical parameters, and the trajectory of these parameters' changes often has a high degree of uncertainty and concealment.

[0008] For example, abnormal fluctuations in current may indicate a potential overload problem, and if this overload continues to develop, it may further cause the line temperature to rise, eventually evolving into a short circuit or fire hazard.

[0009] This chain reaction, from one parameter anomaly to another parameter deterioration, makes it extremely difficult to accurately identify fault transition paths and also increases the complexity of predicting future fault types.

[0010] Therefore, how to accurately capture the signs of a fault evolving from one type to another amidst the dynamic changes of numerous electrical parameters, and to clarify its possible transformation direction and probability, has become a key issue that urgently needs to be addressed in the operation and maintenance of power distribution systems.

[0011] Solving this problem requires not only focusing on the superficial changes in parameters, but also gaining a deeper understanding of the intrinsic connections and evolution logic between different faults, so as to provide maintenance personnel with timely and effective decision-making basis. Summary of the Invention

[0012] This invention provides a method for early warning and analysis of power distribution line faults, mainly including:

[0013] By extracting parameter change sequences from electrical parameters in the power distribution system in real time, performing feature decomposition on the frequency and amplitude range of abnormal fluctuations, analyzing signal interference intensity, and obtaining a preliminary distribution of abnormal patterns;

[0014] Based on the preliminary distribution of abnormal patterns, combined with the duration and time window distribution, the parameter change trends are classified. If the abnormal fluctuation frequency in the classification results exceeds the preset abnormal fluctuation frequency threshold, the potential chain reaction initiation point is determined.

[0015] By identifying potential chain reaction starting points, we obtain the historical failure frequency and environmental impact factors corresponding to the associated equipment identifiers, construct a failure type association dataset, analyze the failure type hierarchy, and obtain the type transformation direction sequence.

[0016] Based on the type conversion direction sequence, the conversion time interval and path weight allocation are integrated to calculate the probability of each conversion path. If a path corresponds to a hidden feature label and the probability is higher than the preset probability threshold, then a high-risk conversion path is determined.

[0017] By identifying high-risk transformation paths, a fault evolution graph is generated. Based on node connection density and node influence radius, graph structure analysis technology is used to sort out path cycle patterns and determine the logical chain of fault evolution.

[0018] Based on the fault evolution logic chain, the impact of parameter changes on the proportion of missing data in the chain is simulated, and risk assessment is carried out in combination with historical fault frequency. If the fluctuation range in the simulation continues to expand, an early warning signal is triggered, and the basis for risk decision-making is obtained.

[0019] By updating the power distribution system monitoring rules based on risk decision-making, adjusting the monitoring thresholds according to environmental impact factors and signal interference intensity, and optimizing the monitoring logic by combining hidden feature markings, a refined fault identification mechanism is obtained.

[0020] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0021] This invention discloses a method for power distribution fault monitoring and risk assessment. Addressing the difficulty in effectively identifying potential cascading reactions, fault type transformations, and high-risk paths caused by abnormal fluctuations when monitoring electrical parameters in real-time power distribution systems, this method extracts parameter change sequences from current and voltage signals and performs feature decomposition. Combined with time-series data processing to analyze interference intensity, it obtains the distribution of abnormal patterns. Then, based on duration and time window classification trends, it determines the starting point of potential cascading reactions. A dataset is constructed by integrating historical fault frequencies and environmental factors. Probabilistic inference is used to analyze type transformation sequences and calculate path probability to identify high-risk paths. A fault evolution map is generated and the logical chain is analyzed. Risk assessment is conducted by simulating the impact of parameter changes, triggering early warnings and updating monitoring rules. This method overcomes the neglect of hidden features and dynamic risks in traditional monitoring, achieving a refined fault identification mechanism, ultimately improving the stability and early warning accuracy of power distribution systems and reducing cascading losses. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for early warning and analysis of power distribution line faults according to the present invention;

[0023] Figure 2 This is a schematic diagram of a method for early warning and analysis of power distribution line faults according to the present invention;

[0024] Figure 3 This is another schematic diagram of a method for early warning and analysis of power distribution line faults according to the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0026] like Figure 1-3 The method for early warning and analysis of power distribution line faults in this embodiment may specifically include:

[0027] S101. By extracting parameter change sequences from electrical parameters in the power distribution system in real time, performing feature decomposition on the abnormal fluctuation frequency and fluctuation amplitude range, analyzing the signal interference intensity, and obtaining the preliminary distribution of abnormal patterns.

[0028] Real-time current and voltage signal data are collected by sensors in the power distribution system. Digital processing technology is used to preprocess the raw data to obtain a preliminary electrical parameter dataset. Change sequence data are extracted from this dataset to identify abnormal fluctuations. A preset threshold range is used for initial screening to determine a set of potential anomaly points. For the selected anomaly point set, the frequency and amplitude ranges of abnormal fluctuations are analyzed. Feature decomposition technology is used to decompose the anomaly points in multiple dimensions, resulting in a set of decomposed feature vectors. Using these feature vectors and time-series processing techniques, the intensity of signal interference is quantitatively analyzed. If the signal interference intensity exceeds a preset threshold, it is identified as a high-risk interference point, resulting in a list of high-risk interference points. Based on this list, the correspondence between signal interference and abnormal patterns is analyzed. Support vector machine (SVM) algorithms are used to classify the abnormal patterns, determining their preliminary distribution categories. For each preliminary distribution category, a corresponding anomaly pattern distribution map is generated. This map data is used to visualize the correlation between abnormal fluctuations and signal interference, revealing the distribution characteristics of the abnormal patterns. By comparing and analyzing the distribution characteristics of abnormal patterns with historical time-series data, if there is a significant deviation between the distribution characteristics and historical abnormal patterns, they are marked as new abnormal patterns, and the potential risk level of the new abnormal patterns is determined.

[0029] In one possible implementation, current and voltage signal data are collected in real time by sensors in the power distribution system. The collected raw data is preprocessed using cleaning and normalization techniques in digital processing technology, such as handling missing values, outliers, and standardization, to obtain a preliminary organized electrical parameter dataset.

[0030] In one possible implementation, the step of extracting change sequence data from the initially compiled electrical parameter dataset, identifying abnormal fluctuations in the change sequences, and performing preliminary screening using a preset threshold range to determine a set of potential anomaly points, wherein the preset threshold range is a dynamic or static standard that can effectively distinguish between normal fluctuations and potential abnormal fluctuations, and its specific feasible implementation is as follows:

[0031] 1. Historical Statistical Analysis Method (Static / Initial Threshold): Based on historical normal operation data of the power distribution system, statistical analysis is performed on the change sequences of current and voltage to calculate their mean (μ) and standard deviation (σ). The initial threshold range can be set as μ ± kσ, where k is a predetermined coefficient (such as 2 or 3, i.e., the 2σ or 3σ principle). Fluctuations exceeding this range are initially screened as an abnormal point set. For example, step S101 requires "identifying abnormal fluctuations," which requires an initial boundary to distinguish between "normal" and "abnormal."

[0032] 2. Dynamic Adaptive Adjustment Method: During system operation, the monitoring rules and thresholds are dynamically adjusted based on the risk decision-making criteria, environmental impact factors, and signal interference intensity obtained in step S107. For example, the threshold range should be appropriately widened during periods of high ambient temperature and strong signal interference; conversely, it should be tightened. As step S107 explicitly states, "the monitoring threshold should be adjusted according to the environmental impact factors and signal interference intensity." This indicates that the threshold is not fixed but dynamically optimized.

[0033] 3. Optimization method based on fault evolution model: A fault type association dataset and type transformation direction sequence are constructed using Bayesian inference with equal probability. These models provide feedback for thresholds: if a specific parameter change (such as a small but continuous increase in temperature) is identified as a high-risk transformation direction in the model, even if its fluctuation amplitude is still within 3σ, its corresponding threshold should be lowered to identify it as an anomaly earlier. For example, steps S103 and S104 are used to generate high-risk transformation paths, and step S106 is used to "simulate parameter changes" for risk assessment. These analysis results are ultimately fed back to S107 for "updating power distribution system monitoring rules" and "dynamically adjusting monitoring thresholds".

[0034] In one possible implementation, for the selected set of anomaly points, the frequency and amplitude ranges of the anomalous fluctuations are analyzed to determine the original dimensions to be decomposed, such as the frequency, amplitude, and duration of the anomalous fluctuations. A suitable decomposition technique for the time-series data and multi-dimensional features is selected, such as: 1. Fast Fourier Transform (FFT): suitable for decomposing time-domain signals into the frequency domain to obtain the main anomalous frequency components; 2. Wavelet decomposition: suitable for simultaneous multi-scale analysis in the time and frequency domains, effectively capturing transient and local anomalies; 3. Principal Component Analysis (PCA): suitable for dimensionality reduction and extracting the most representative main feature directions (feature vectors) of the anomalous fluctuations. The original anomalous fluctuation data is input into the selected decomposition algorithm to perform mathematical transformations and decomposition of the signal. The most representative values ​​from the decomposition results are selected as feature vectors. These vectors are used to quantify the core attributes of the anomalous fluctuations, such as the amplitude of the main frequency components and the energy proportion after decomposition.

[0035] The intensity of signal interference is quantitatively analyzed by combining the decomposed feature vector group with time-series processing techniques. If the signal interference intensity exceeds a preset signal interference intensity threshold, it is determined to be a high-risk interference point, and a list of high-risk interference points is obtained. The specific steps of the quantitative analysis are as follows:

[0036] 1. Interference Feature Extraction and Isolation: Identifying and isolating feature components related to signal interference from the feature vector set. For example, after frequency domain decomposition (a form of feature decomposition), high-frequency, random feature components are often related to environmental noise or signal interference. The goal is to ensure that the quantitative analysis targets only the interference signal, rather than the inherent characteristics of the fault itself.

[0037] 2. Temporal Accumulation and Smoothing: Temporal processing techniques (such as moving average, exponential smoothing, or Kalman filtering) are used to accumulate or smooth the isolated interference characteristic components over time. The aim is to eliminate instantaneous interference spikes, obtain the persistence and stability trend of interference intensity, and make it more representative.

[0038] 3. Interference Energy or Power Calculation: Based on the results of time-series processing, calculate the energy or average power of the interference signal within a specific time window. For example, calculate the norm of the interference eigenvector (...). The goal is to transform a complex eigenvector into a single, comparable quantified value, namely, the "intensity of signal interference".

[0039] 4. Threshold Comparison and Judgment: The calculated quantized value of signal interference intensity is compared with a preset signal interference intensity threshold. If the signal interference intensity exceeds the preset threshold, it is judged as a high-risk interference point and added to the high-risk interference point list. The purpose is to complete the risk assessment and distinguish between acceptable interference and high-risk interference that requires attention.

[0040] The specific method for setting the preset signal interference threshold is as follows:

[0041] 1. Determining the initial threshold (baseline setting)

[0042] Method: Historical baseline method

[0043] • Steps: Collect signal interference data of the power distribution system under long-term normal operation.

[0044] • Perform statistical analysis on these historical data to calculate the mean (μ) and standard deviation (σ) of the interference intensity.

[0045] • The initial threshold is set to μ plus an empirical safety margin, for example: initial threshold = μ + kσ (where k can be 2 or 3, i.e., the 2σ or 3σ principle).

[0046] • Function: To ensure that the system can identify interference that significantly exceeds historical normal levels when it is first put into use.

[0047] 2. Dynamic adaptive adjustment (optimization mechanism)

[0048] The core of this method lies in the dynamic updating of the threshold to cope with the ever-changing operating environment.

[0049] • Based on environmental impact factors: High ambient temperature, high humidity or high salt spray and other environmental impact factors will naturally increase signal attenuation and noise.

[0050] Adjustment: When environmental factors deteriorate, appropriately increase the interference intensity threshold to prevent oversensitivity.

[0051] • Based on real-time signal interference intensity: When the real-time acquired signal interference intensity continues to rise over a period of time, but has not yet reached the current threshold.

[0052] Adjustment: The trend is calculated using time-series processing techniques such as exponential weighted average or sliding window average. If the disturbance trend is significantly upward, the threshold is dynamically increased to adapt to the new normal.

[0053] • Based on hidden feature markers: If quantitative analysis reveals that a specific interference pattern, although not strong, is highly correlated with hidden feature markers, it indicates potential fault evolution.

[0054] Adjustment: The threshold for this specific type of interference has been significantly lowered (tightened) to enable early detection of "hidden faults".

[0055] In one possible implementation, the step of generating a corresponding abnormal pattern distribution map based on the preliminary distribution categories of abnormal patterns, and visually mapping the correlation between abnormal fluctuations and signal interference through the map data to obtain the distribution characteristics of abnormal patterns, is implemented in the following specific way:

[0056] The generation of the map revolves around visualizing mapping relationships:

[0057] 1. Data input: Data from the preliminary distribution categories of anomaly patterns (determined by support vector machine algorithm classification) and the list of high-risk disturbance points (determined by quantitative analysis).

[0058] 2. Constructing the graph structure:

[0059] Different anomaly pattern categories are used as nodes in the graph.

[0060] • Treat signal interference intensity or high-risk interference points as one or more special nodes or attribute dimensions in the graph.

[0061] 3. Visualize mapping relationships:

[0062] • Determine the degree of correlation between anomalous pattern nodes and signal interference through calculation or preset rules (e.g., the frequency of a certain anomalous pattern occurring in a strong interference environment).

[0063] • Connect these relationships using edges in the graph, and use the weight, color, or thickness of the edges to represent the strength of the relationship.

[0064] 4. Final output: Generate a visualization chart (i.e., anomaly pattern distribution map) to intuitively show the distribution of various anomaly patterns under different disturbance conditions.

[0065] The distribution characteristics of the anomalous patterns are the core output of this spectral analysis, aiming to identify the deep correlation between anomalous fluctuations and signal interference. These characteristics include:

[0066] 1. Dominant anomaly pattern: The anomaly pattern category that occurs most frequently or accounts for the largest proportion in the entire spectrum or within a specific area of ​​interference intensity.

[0067] 2. Interference Sensitivity: Identify the abnormal patterns most strongly associated with nodes experiencing high signal interference. These patterns may be false alarms or potential faults most easily triggered under strong interference.

[0068] 3. Temporal-spatial distribution: By combining the collected time-series data and device identifiers, analyze whether the abnormal patterns tend to be concentrated in specific time windows (such as late night or peak hours) or specific device areas (spatial distribution).

[0069] 4. New Anomaly Pattern Labeling: This involves comparing the distribution of map features with historical anomaly patterns. If the current distribution features displayed by Tuno deviate significantly from historical records, these features may be labeled as new anomaly patterns.

[0070] By acquiring these distribution characteristics, the system can determine the potential risk level of novel abnormal patterns, providing key input for the subsequent determination of potential chain reaction initiation points (S102).

[0071] In one possible implementation, statistical or machine learning methods are used to quantify the degree of difference between the current distribution map of abnormal patterns and historical maps. If the distribution characteristics deviate significantly from historical abnormal patterns, they are marked as novel abnormal patterns. The potential risk level of the novel abnormal patterns is then determined. The criteria for determining the significant deviation are as follows:

[0072] 1. Distance metrics (statistics): Calculate the distance between the current distribution features of anomaly patterns (such as feature vectors) and the average features of historical anomaly patterns. Commonly used distance metrics include: Euclidean distance, Mahalanobis distance, or cosine similarity.

[0073] If the distance exceeds a preset statistical threshold (such as 2σ or 3o of the historical distribution variability), it is judged as a significant deviation.

[0074] 2. Probability Distribution Difference (Machine Learning): Compare the probability distribution of each anomaly pattern category in the current map with the distribution of historical records. Common methods include calculating relative entropy (Kullback-Leibler divergence) or JS divergence (Jensen-Shannon divergence).

[0075] If the divergence values ​​of the two distributions are higher than a certain threshold, it indicates that the current distribution is very similar to the historical distribution, that is, there is a significant bias.

[0076] 3. New features / nodes appear: If an anomalous pattern category (new node) that has never been identified in history appears in the graph, or the correlation (edge) between anomalous fluctuations and signal interference is completely different from the historical record.

[0077] This structural change itself constitutes a significant deviation.

[0078] If the distribution characteristics deviate significantly from historical anomaly patterns, it is marked as a new anomaly pattern. Once marked as a new anomaly model, the assessment of its potential risk level usually requires combining its intrinsic characteristics and external correlations. The specific methods for assessing the potential risk level are as follows:

[0079] 1. Correlation strength

[0080] Determine the correlation strength between the new anomaly pattern and the high-risk interference point: If the correlation weight between the new pattern and the high-risk interference signal is extremely high, it indicates that it is a pattern that is sensitive to the external environment and unstable, with a high risk level.

[0081] Basis: Step S101 analyzes high-risk interference points and uses support vector machine to classify abnormal patterns.

[0082] 2. Fluctuation amplitude / frequency

[0083] Analyze the fluctuation frequency and amplitude under the new mode: If the abnormal fluctuation frequency or amplitude range measured under the new mode is close to or exceeds the fault threshold, the risk level is high.

[0084] Basis: Both steps S101 and S102 mention the analysis of the frequency and amplitude range of abnormal fluctuations.

[0085] 3. Duration trend

[0086] Combining duration and time window distribution: If a new anomaly pattern persists within a long time window and the fluctuation frequency continues to exceed the threshold, it indicates a potential chain reaction and a high risk level.

[0087] Basis: Step S102 is explicitly used to determine the starting point of a potential chain reaction.

[0088] 4. Historical Fault Correlation

[0089] The risk level is high if the new anomaly pattern is associated with a device that has historically experienced frequent or significant failures.

[0090] Basis: Step S103 obtains the historical failure frequency through the potential cascading reaction starting point.

[0091] By taking into account the above multi-dimensional factors, the system can assign a risk level (e.g., Level I / Minor, Level II / Moderate, Level III / Severe) to new abnormal patterns, so that different monitoring and early warning strategies can be adopted subsequently.

[0092] S102. Based on the preliminary distribution of abnormal patterns, combined with the duration and time window distribution, classify the parameter change trends. If the abnormal fluctuation frequency in the classification results exceeds the preset abnormal fluctuation frequency threshold, then determine the potential chain reaction starting point.

[0093] The system acquires abnormal pattern data recorded in the system, organizes preliminary distribution characteristics, and uses data grouping techniques to initially divide the abnormal patterns according to time windows, resulting in grouped abnormal datasets. From these grouped datasets, information related to duration and time windows is extracted, and parameter changes within each group are analyzed to determine the trend characteristics. Based on these trend characteristics, the data within each group is classified, generating classification result data. Frequency statistics are performed on the abnormal fluctuations in the classification results to obtain fluctuation frequency values. If the fluctuation frequency value exceeds a preset abnormal fluctuation frequency threshold, the group is identified as a potential chain reaction initiation point and marked as a high-risk group. Abnormal patterns and parameter change information are extracted from the data in the high-risk group, and a support vector machine algorithm is used for further pattern analysis to determine possible trigger points for chain reactions. Based on the possible trigger points, corresponding monitoring strategies are generated for the time windows and durations, and data related to the trigger points is tracked in real time to obtain dynamic data on abnormal fluctuations. By dynamically changing data, the calculation results of the fluctuation frequency are continuously updated. If the updated fluctuation frequency still exceeds the preset abnormal fluctuation frequency threshold, the exact location of the chain reaction starting point is confirmed, and corresponding early warning information is generated.

[0094] If the fluctuation frequency value exceeds the preset abnormal fluctuation frequency threshold, it is determined that there is a potential chain reaction starting point in the group and it is marked as a high-risk group. The implementation of the preset abnormal fluctuation frequency threshold can be divided into two stages: initial setting and dynamic optimization, so as to achieve the purpose of accurately identifying potential chain reaction starting points.

[0095] 1. Setting the initial threshold (based on historical data)

[0096] In the initial stage of system operation, this threshold value can be determined based on the results of historical data analysis:

[0097] Historical statistical analysis method:

[0098] • Steps: Collect abnormal fluctuation frequency data of the power distribution system under long-term normal operation or known non-chain reaction anomalies.

[0099] Calculate the mean (μ) and standard deviation (σ) of these historical frequency values.

[0100] • The initial threshold is set as: initial threshold = μ + kσ (where k is an empirical safety factor, such as 2 or 3).

[0101] • Function: To ensure that only those frequency values ​​that are significantly higher than normal abnormal fluctuation levels are identified as potential cascading reaction initiations.

[0102] 2. Optimization of dynamic thresholds (based on risk feedback)

[0103] Once the system is up and running, this threshold needs to be dynamically adjusted based on the results of subsequent steps (especially risk assessment and monitoring rule updates) to improve the accuracy of early warnings.

[0104] • Optimization Basis: Risk Decision Basis

[0105] Based on the risk assessment results: if, after simulating parameter changes in S106, it is found that even a slight fluctuation at a certain frequency can easily lead to a continuous increase in the fluctuation amplitude and trigger a warning signal, then that threshold needs to be tightened (lowered).

[0106] Based on the fault evolution logic chain, the impact of parameter changes on the proportion of missing data in the chain is simulated. Risk assessment is conducted in conjunction with historical fault frequencies. If the fluctuation range in the simulation continues to expand, an early warning signal is triggered, providing a basis for risk decision-making.

[0107] • Optimization basis: Refined fault identification mechanism

[0108] Adjustments based on environment and interference: S107 explicitly states that monitoring thresholds should be adjusted based on environmental impact factors and signal interference intensity.

[0109] Adjustment: When environmental influencing factors (such as high humidity, high temperature) or signal interference intensity are high, the threshold may be appropriately relaxed (raised) because the misjudgment rate may increase; conversely, it should be tightened (lowered).

[0110] • Optimization basis: Support Vector Machine (SVM) analysis

[0111] Based on the pattern analysis results: Step S102 uses Support Vector Machine (SVM) to perform pattern analysis on the data within the high-risk group to determine the possible trigger points of the chain reaction.

[0112] Adjustment: If SVM analysis determines that a specific abnormal pattern type is a high-probability trigger point for a chain reaction, the abnormal fluctuation frequency threshold corresponding to that specific pattern should be lowered to achieve more sensitive early warning. S103. By identifying the potential cascading reaction starting point, obtain the historical failure frequency and environmental impact factors corresponding to the associated equipment identifier, construct a failure type association dataset, analyze the failure type hierarchy, and obtain the type transformation direction sequence.

[0113] By extracting data from the equipment identification database, historical fault records and fault frequency information related to the initiation point of a chain reaction are obtained to determine a preliminary fault distribution pattern. Based on the preliminary fault distribution pattern, combined with environmental impact factor data, the correlation between environmental impact and fault frequency is analyzed to obtain the potential weight of the environment on fault types. Using Bayesian inference methods, an association dataset is constructed for fault types and environmental impact weights to derive the hierarchical structure of fault types. Through the hierarchical structure of fault types, the transformation direction between types is analyzed. For example, a low-level fault may escalate to a higher-level fault. The probability distribution of transformation directions is obtained to determine the main transformation paths. Based on the transformation path results and combined with historical fault data, if the transformation probability of a certain fault type is higher than a preset transformation probability threshold, it is marked as a high-risk transformation direction, and risk warning information is output. Data on high-risk transformation directions is obtained, and their correspondence with equipment identification is analyzed to generate equipment-level fault prevention strategies and obtain targeted prevention and control measures. Through the implementation effect data of prevention and control measures, the fault type hierarchy and transformation direction sequence in the association dataset are updated to continuously optimize the accuracy of subsequent analyses.

[0114] Based on the results of the conversion path and combined with historical fault data, if the conversion probability of a certain fault type is higher than a preset conversion probability threshold, it is marked as a high-risk conversion direction, and a risk warning message is output. The preset conversion probability threshold is set as follows:

[0115] 1. Setting the initial threshold (based on historical experience)

[0116] In the initial stage of system operation, this threshold can be determined based on historical fault data and expert experience:

[0117] Historical statistical method:

[0118] • Steps: Analyze the failure type transformation paths that have historically led to major failures or cascading losses, and calculate their average transformation probability.

[0119] • The initial threshold is set to a value higher than the historical average probability. For example: initial threshold = historical average conversion probability + safety margin.

[0120] • Function: To ensure that only those paths with a conversion probability significantly higher than the historical average are marked as high-risk conversion directions.

[0121] • Expert experience method:

[0122] • Based on the maintenance experts' understanding of specific fault types (such as overload, insulation aging), directly set an experience rate value (such as 70% or 80%) as an initial threshold. If it exceeds this value, it needs to be addressed immediately.

[0123] 2. Optimization of dynamic threshold (based on system feedback)

[0124] Once the system is operational, this threshold needs to be dynamically adjusted based on subsequent analysis and feedback to improve sensitivity to actual risks.

[0125] • Optimization basis: Environmental impact factors

[0126] Based on the potential impact weight of the environment on the fault: Step S103 analyzes the correlation between environmental impact and fault frequency, and obtains the potential impact weight of the environment on the fault type.

[0127] Adjustment: When the weight of environmental impact factors (such as severe weather or sustained high temperatures) on the transformation of a certain fault increases, the threshold should be tightened (lowered) even if the transformation probability is slightly lower, so as to mark the path as high risk earlier.

[0128] • Optimization basis: Hidden feature labeling

[0129] Based on the concealment of the transformation path: S104 mentions if a certain path corresponds to a concealed feature marker.

[0130] Adjustment: If the probability of a certain transformation direction is not high, but it is associated with hidden features, meaning that it is difficult to detect once it occurs, the threshold should be tightened (lowered) to ensure that such hidden risks are identified.

[0131] • Optimization basis: The effectiveness of prevention and control measures

[0132] Based on feedback on the actual intervention effect: S103 mentioned that the fault type hierarchy and transformation direction sequence in the associated dataset should be updated through the data on the implementation effect of prevention and control measures.

[0133] Adjustment: If the actual incidence of a high-risk conversion direction decreases significantly after preventive measures are taken, the threshold can be appropriately relaxed (raised); otherwise, it should be tightened.

[0134] Through this dynamic mechanism, the threshold can be self-optimized as the operating environment of the power distribution system, the effectiveness of prevention and control measures, and the fault conversion mode change, making risk warning more accurate.

[0135] S104. Based on the type conversion direction sequence, the conversion time interval and path weight allocation are integrated to calculate the probability of each conversion path. If a path corresponds to a hidden feature marker and the probability is higher than the preset probability threshold, then a high-risk conversion path is determined.

[0136] The process begins by acquiring data records of conversion types and sequence directions. Sequence information for each conversion path is extracted and categorized to obtain a preliminary path classification set. Based on this set, corresponding interval time data is acquired. Combining path weights and weight allocation rules, the interval time of each path is weighted to obtain a time-weighted result. Using this time-weighted result, the probability value of each conversion path is calculated. A pre-established probability model is used for probability assessment to determine the probability value record for each path. For each probability value record, hidden features and feature markers are acquired. If a path's hidden features match its feature markers, it is marked as a potential risk path. Based on the probability value records of potential risk paths, if a path's probability value exceeds a preset probability threshold, it is identified as a high-risk path, and a high-risk path list is output. Using this high-risk path list, detailed data on relevant conversion types and sequence directions is obtained, generating path risk correlation analysis records to determine the final risk path distribution information.

[0137] Based on the preliminary path classification set, corresponding interval time data is obtained. Combining path weights and weight allocation rules, the interval time of each path is weighted to obtain the time-weighted result for each path. The specific weight allocation rules are as follows:

[0138] 1. Initial weight allocation (based on risk and frequency)

[0139] • Weighting factors: High-risk conversion direction, whether the path belongs to the high-risk conversion direction marked in step S103.

[0140] Weighting rules: Assign higher weights: If the starting or ending point of the path is a high-risk fault type, or the path itself is a high-risk transformation direction, a higher weight should be assigned.

[0141] • Weighting factors: Historical failure frequency, the frequency with which the devices or failure types involved in the path have occurred in history.

[0142] Weighting rules: Assign high weights: Paths with high historical frequency should have higher weights because they have a greater possibility of being transformed again and have a greater impact.

[0143] • Weighting factors: Hidden feature labels, whether the path corresponds to a hidden feature label.

[0144] Weighting rules: Assign high weights: Paths with high concealment are difficult to detect once they occur, even if the probability or frequency is not high. Therefore, they should be assigned higher risk weights.

[0145] 2. Weighted processing rules (merging time and weight)

[0146] The weighting rules should use mathematical methods to integrate path weights with conversion time intervals to obtain the final time-weighted result.

[0147] • Objective: High-risk conversions completed in a short period of time should have a higher probability of success in the end.

[0148] • Example rules (formula deduction):

[0149]

[0150] Alternatively, an exponential decay model can be used to process the interval time.

[0151]

[0152] in, It is a decay coefficient that ensures that the longer the conversion time interval, the smaller its contribution to the final "possibility".

[0153] This weighted processing ensures that the importance (weight) and urgency (time interval) of the path are both included in the probability calculation, making the subsequent risk assessment more accurate.

[0154] In one possible implementation, the step of calculating the probability value of each transformation path using time-weighted results, evaluating the probability using a pre-established probability model, and determining the probability value record for each path is as follows:

[0155] For example,

[0156] 1. Markov chain model

[0157] Structural features:

[0158] • Nodes (Status): Each node in the diagram represents a fault type (e.g., overload, insulation aging, loose connection, etc.).

[0159] • Edge (transition): Directed edges between nodes represent the direction of fault type transition (e.g., from "overload" to "insulation aging").

[0160] • Core parameter: Transition probability. Each edge carries a probability value for transitioning from one state to the next (i.e., the transition probability obtained by S103).

[0161] • Evaluation input: The time-weighted result calculated in step S104.

[0162] How to conduct a probability assessment:

[0163] The model combines the transformation probability (S103 result) and the time-weighted result (S104 result) to calculate the cumulative probability of a complete transformation path (consisting of multiple consecutive transitions) occurring.

[0164] The probability value of a path = f(the conversion probability on the path x the time-weighted result of the path)

[0165] ·f(·) may be a product function or some kind of cumulative function.

[0166] • Function: To evaluate paths with high conversion probability and short conversion time intervals (high time-weighted results), resulting in higher probability values.

[0167] 2. Bayesian network model

[0168] Structural features:

[0169] Nodes: In addition to fault type status, nodes can also include environmental impact factors, signal interference intensity, equipment aging degree, etc. as condition variables.

[0170] Edges: Directed edges between nodes represent causal relationships or conditional dependencies (e.g., "high temperature" increases the probability of "insulation aging").

[0171] Core parameter: Conditional Probability Table (CPT). Defines the probability of a child node occurring given the state of the parent node.

[0172] How to conduct a probability assessment:

[0173] The model constructs a network using the fault type association dataset and transformation direction sequence determined in S103. When assessing probability, the model combines real-time monitored environmental, disturbance, and time-weighted results with Bayesian inference to dynamically calculate the posterior probability (i.e., probability value) of a transformation path occurring under current monitoring conditions.

[0174] In one possible implementation, for each probability value record, the hidden features and feature label information of each path are obtained. If the hidden features of a path match the feature label, it is marked as a potential risk path. The hidden features refer to the inherent attributes of the failure evolution path itself that are not easily detected by conventional monitoring methods, such as small-amplitude slow characteristics, high randomness characteristics, multi-dimensional coupling characteristics, and specific spatiotemporal characteristics. The feature label information refers to tags or identifiers in a pre-established database used to identify and label the aforementioned hidden features. The specific matching method for marking a path as a potential risk path if its hidden features match the feature label is as follows:

[0175] 1. Matching input data

[0176] •Hidden feature vectors ( ): Quantitative features extracted from path parameter fluctuation data, for example:

[0177] · : The average value of the fluctuation range (the smaller the value, the more concealed the fluctuation).

[0178] · Duration length (the longer the duration, the more concealed).

[0179] · Signal randomness / entropy (the higher the value, the more concealed the signal).

[0180] • Feature labeling information ( ): Pre-defined concealment criteria in the database, for example:

[0181] · Minimum fluctuation threshold (values ​​below this are hidden).

[0182] · Detection difficulty score (a score higher than this indicates high risk).

[0183] · Hidden fault type label (whether it contains a specific label)

[0184] 2. Feasible matching algorithms

[0185] A. Rule-based logical matching: This is the most direct matching method, which determines whether a match is found by setting a series of logical conditions.

[0186] Algorithm structure:

[0187]

[0188] Example rules:

[0189] Rule 1 (Slow Feature Matching):

[0190]

[0191] Rule 2 (High-risk score matching):

[0192]

[0193] B. Similarity-based matching algorithm, which is suitable for cases where the hidden features are multi-dimensional vectors, calculates the similarity between the hidden feature vector of the current path and the known hidden pattern vectors in the hidden fault database.

[0194] Algorithm structure:

[0195]

[0196] Algorithm selection: Cosine similarity or weighted Euclidean distance can be used.

[0197] Matching determination:

[0198]

[0199] or

[0200]

[0201] This method is more flexible and can identify new anomalies that are “similar” to historical hidden fault patterns.

[0202] C. Final determination of paths marked as potential risks

[0203] After a successful match, the path must meet the final judgment condition (S104 Original Description):

[0204]

[0205] Only when a path is both concealed and has a high probability of occurrence is it ultimately identified as a high-risk conversion path.

[0206] The method involves recording the probability values ​​of potential risk paths. If the probability value of a path exceeds a preset probability threshold, it is determined to be a high-risk path, and a list of high-risk paths is output. Since the probability value is the result of a fusion calculation of the conversion probability (S103) and the time-weighted result (S104), the preset probability threshold must be dynamic and based on historical risks. The specific setting method is as follows:

[0207] 1. Setting the initial threshold (based on historical risk)

[0208] Method: Risk-driven approach

[0209] • Steps: Trace the failure transition paths in history that led to serious consequences or system outages.

[0210] • Record the range of probability values ​​for these historically high-risk paths before conversion.

[0211] • The initial threshold is set slightly below the lowest probability value of historical high-risk paths to ensure that all historical high-risk events are captured.

[0212] For example: Initial threshold = historical lowest high-risk probability value - safety margin.

[0213] • Function: To establish a baseline and ensure that the system has the ability to provide early warnings of known high-risk scenarios.

[0214] 2. Optimization of dynamic thresholds (based on early warning targets)

[0215] This threshold will be dynamically adjusted based on the system's warning targets and feedback, specifically:

[0216] • Optimization basis: Early warning window requirements

[0217] If the system requires an earlier warning time (i.e. a longer warning window).

[0218] Adjustment: The threshold needs to be tightened (lowered). A lower probability value can also trigger an alert, thus allowing more time for intervention.

[0219] • Optimization basis: False alarm rate feedback

[0220] If the system generates a large number of false alarms during operation (marking non-risk paths as high-risk).

[0221] Adjustment: The threshold needs to be relaxed (raised). Only paths with a higher probability (closer to the actual risk) should trigger alerts, reducing the operational burden.

[0222] • Optimization basis: Hidden feature association

[0223] If the current path is marked as a hidden feature path and its risk is extremely high.

[0224] Adjustment: For concealed feature paths, the threshold should be significantly tightened (lowered) to ensure that concealed risks are identified first, even if their probability value is not high.

[0225] In summary, the preferred implementation method is to first set a risk baseline as an initial value based on historical major failure events, and then dynamically and adaptively optimize and adjust this value according to the system's early warning requirements, false alarm situation, and matching results of hidden characteristics.

[0226] S105. Through high-risk transformation paths, generate a fault evolution map. Based on node connection density and node influence radius, use graph structure analysis technology to sort out the path cycle pattern and determine the fault evolution logic chain.

[0227] By using a pre-established fault data archive, relevant records of high-risk transformation paths are obtained. Key nodes and transformation events in these records are categorized and labeled to obtain a preliminary fault evolution graph framework. Based on this framework, graph structure analysis is employed to perform a depth-first traversal of the connections between nodes, determining the distribution pattern of node connection density. The distribution pattern of node connection density is used to analyze the weight of each node in the fault evolution graph. A weighted calculation method is used to determine the triggering conditions of key nodes, resulting in an initial sequence of fault logic chains. For this initial sequence, feature data of path loop patterns is obtained. If duplicate loop paths exist, they are merged to determine the optimized loop pattern identification result. Based on the optimized loop pattern identification result, potential risk points in high-risk transformation paths are identified. A traversal comparison method is used to determine the correlation strength between risk points and the fault logic chain. Based on this correlation strength, a structural framework for the evolution chain is constructed. Using this framework, key links in the transformation paths are analyzed. If the trigger probability of a key link is higher than a preset trigger probability threshold, it is prioritized to determine the final fault evolution graph. Based on the final fault evolution map, a logical closed loop of high-risk transformation path is generated. The weak nodes in the closed loop are verified for correlation, for example, by backtracking historical data, simulation, or physical experiments, to obtain a complete fault evolution logical chain.

[0228] The process of obtaining relevant records of high-risk transformation paths through a pre-established fault data archive, classifying and labeling key nodes and transformation events in the records to obtain a preliminary fault evolution map framework, involves a structured and knowledge-based process that integrates historical data of the power distribution system, environmental information, and the inherent logic between faults. This process primarily relies on the results of the aforementioned steps (knowledge accumulation prior to S103 and S104) of this application, specifically:

[0229] 1. Data Sources and Collection

[0230] The foundation for establishing the archives is historical failure data and operating environment data:

[0231] • Historical Fault Records: Collects every fault event in the history of the power distribution system, including:

[0232] • Fault type (e.g., overload, short circuit, insulation aging).

[0233] • Time of occurrence and duration of the fault.

[0234] • Associated device identifier (for S105 / Step 1).

[0235] • Fault recovery measures and results.

[0236] • Environmental impact factors: Collect environmental data related to the occurrence of the failure, such as temperature, humidity, wind speed, pollution level, etc.

[0237] 2. Knowledge Structuring and Connection (S103 Outcome)

[0238] At the heart of the archives are structured data and knowledge, which is primarily achieved through the steps described in S103:

[0239] • Fault type associated dataset:

[0240] Based on historical failure frequency and device identification, establish correlations between failure types. Record which failures are more likely to occur on which devices, and the recurrence of these failures.

[0241] • Type conversion direction sequence:

[0242] Probabilistic inference techniques (such as Bayesian inference) are employed to analyze the evolutionary relationships between historical fault types (e.g., the transformation from "overload" to "insulation aging"). Recording the probability and direction of the transformation between different fault types forms the basis for constructing the logical chain.

[0243] • Historical failure frequency record:

[0244] Frequency statistics are performed for each device identifier and fault type. This provides input for subsequent risk assessment (S106).

[0245] 3. Risk and Characteristic Labeling (S104 Results)

[0246] The archives should also include expert knowledge and labeling information for identifying high-risk and difficult-to-detect pathways:

[0247] • Hidden feature labeling information: Store fault feature patterns that do not fluctuate significantly and are difficult to detect by traditional methods (e.g., thresholds that change slowly and slightly, judgment criteria with high randomness).

[0248] • Weighting rules: Stores the rules for assigning weights to conversion paths based on risk level and historical frequency (used for weighting in S104).

[0249] Summarize

[0250] Therefore, the pre-established fault data archive is not a single data file, but a comprehensive historical database and analytical knowledge base. It is a structured information set that is ultimately formed by a series of operations such as classifying, statistically analyzing, probabilistically inferring, and labeling historical operation and fault data, and is used by subsequent steps of the system (such as generating fault logic chains).

[0251] In one possible implementation, the distribution pattern of node connection density is used to analyze the weight of each node in the fault evolution graph. A weighted calculation method is then used to determine the triggering conditions of key nodes, resulting in an initial sequence of the fault logic chain. The key nodes refer to the fault type nodes in the fault evolution graph that, once activated (i.e., a fault occurs), are most likely to cause a chain reaction or system collapse. The triggering conditions refer to the conditions that must be met for a node to transition from a "normal" or "low-risk" state to an "active" or "faulted" state.

[0252] Trigger condition type:

[0253] 1. Impact Weight Threshold: The impact weight of a node (derived through node connection density analysis) must be higher than a preset critical weight threshold. S105 explicitly states that it is necessary to "analyze the impact weight of each node in the fault evolution map," and only nodes with high weights should be considered "critical nodes."

[0254] 2. High Probability Value Association: The node must be the starting point or a key transition point on a high-risk transformation path (determined in S104), and the probability value of this path must meet the preset probability threshold. The result of S104 is the input of S105, and nodes on high-risk paths are the priority triggering conditions.

[0255] 3. Actual parameter changes: Abnormal fluctuations in the actual electrical parameters (such as current and voltage) corresponding to the node have been continuously occurring, and the amplitude or frequency of these fluctuations exceeds the local threshold (further application of S101). The essence of fault warning requires parameter changes, and the actual parameter changes are the final confirmation signal that triggers the warning.

[0256] 4. Low data missing correlation: The data missing ratio corresponding to a node must be lower than a preset tolerance threshold (data missing ratio mentioned in S106). If the data missing ratio of a node is too high, its monitoring results are unreliable and cannot be easily judged as a "triggered" state.

[0257] Therefore, the triggering condition of a critical node is the result of a multi-dimensional comprehensive judgment. It requires that the node itself has high importance (weight) in the graph structure, and that actual monitoring data supports the occurrence of its failure.

[0258] The initial sequence of the fault logic chain refers to a set of main paths extracted from the graph, consisting of key nodes and high-risk transformation directions between them. The specific steps to obtain the initial sequence are as follows:

[0259] 1. Determine the set of key nodes:

[0260] Based on the triggering conditions of the above key nodes, all nodes that meet the conditions are selected from the fault evolution graph to form a key node set.

[0261] 2. Extract high-weight edges:

[0262] • Based on the edges (transformation directions) connecting key nodes in the graph, select those edges with high transformation probabilities or high time-weighted results (i.e., high-risk transformation paths).

[0263] 3. Construct the initial path:

[0264] • Search algorithm: Graph search algorithms such as depth-first search (DFS) or breadth-first search (BFS) are used.

[0265] • Search target: Starting from the initial fault source node (such as the potential chain reaction starting point determined by S102), traverse and connect all critical nodes along the high-weight edges until the termination node (such as the fault type that caused the system interruption) is reached.

[0266] 4. Generate the initial sequence:

[0267] Arrange the main paths containing key nodes obtained from the search according to time or transformation order to form the initial sequence of the fault logic chain (i.e. a set of high-risk, high-probability evolution paths).

[0268] This initial sequence forms the basis for subsequent steps (steps 4-7 of S105) to perform path optimization, loop pattern recognition, and logical loop closure verification.

[0269] The structure framework of the evolution chain is used to analyze key links in the transformation path. If the trigger probability of a key link is higher than a preset trigger probability threshold, it is prioritized to determine the final fault evolution map. The preset trigger probability threshold is set as follows:

[0270] 1. Setting the initial threshold (based on historical risk and criticality)

[0271] Method: Historical Key Nodes Method

[0272] • Steps: Trace historical cases of failures that led to major system outages or cascading effects.

[0273] • Statistics show the range of recorded values ​​for the trigger probability of key links (such as the starting point of a chain reaction or a key transformation point) when they transform into the next type of failure in these cases.

[0274] • Initial threshold setting: Take the lowest trigger probability value among historical high-risk cases, and then subtract a small safety margin.

[0275] Initial threshold = Lowest historical trigger probability of critical events - Safety margin

[0276] • Function: To ensure that the system can capture all known, highly dangerous critical process transformations in history.

[0277] 2. Optimization of dynamic thresholds (based on risk feedback and system requirements)

[0278] This threshold must be dynamically adjusted in conjunction with subsequent steps (especially the optimization mechanism in S107).

[0279] • Optimization criteria such as early warning accuracy requirements: If the system requires extremely high early warning accuracy, in order to reduce false alarms (false alarms will consume operation and maintenance resources).

[0280] Adjustment: Threshold relaxed (raised). Only steps with an extremely high conversion rate (e.g., over 90%) will be marked as priority to ensure resources are focused on the most urgent events.

[0281] • Optimization is based on feedback from the logical closed-loop verification: Step S107: For the logical closed loop of high-risk conversion paths, verify the weak points within it. If a key link is identified as a "weak point," its triggering will cause the logical closed loop to complete rapidly.

[0282] Adjustment: Tighten (lower) the threshold. The trigger probability threshold for weak nodes should be significantly reduced to ensure that they receive priority labeling even if the probability is slightly lower.

[0283] • Optimization criteria such as environmental / interference factors: Step S107: Adjust the monitoring threshold based on environmental impact factors and signal interference intensity.

[0284] Adjustment: When the system operating environment is harsh, which reduces the reliability of monitoring data, the priority threshold can be appropriately relaxed (raised) to avoid excessive warnings based on uncertain data.

[0285] • Optimization criteria such as hidden feature association: If a key link is highly associated with the hidden feature marker (S104), its priority must be increased.

[0286] Adjustment: Threshold tightened (lowered). This ensures that these hard-to-detect critical risks are prioritized for identification.

[0287] In one possible implementation, the logical closed loop for generating high-risk transformation paths based on the final fault evolution graph refers to a series of fault types sequentially transforming into each other in the fault evolution graph, ultimately returning to the initial fault type, forming a circular chain. This is crucial in fault analysis because a closed loop implies that the risk is self-sustaining or continuously escalating. The specific implementation method for generating the closed loop is as follows:

[0288] 1. Input data: The final fault evolution map (output of step six, S105, including priority marking and refined path).

[0289] 2. Loop closure detection algorithm: Employs loop detection algorithms from graph theory, for example:

[0290] • Depth-first search (DFS): When traversing a graph, if starting from a node, after a series of transformations, a node that has already been visited and is still in the current search path stack is encountered again, a closed loop is formed.

[0291] • Finding Strongly Connected Components (SCCs): If some nodes in the graph form an SCC, then there are one or more cyclic paths between them.

[0292] 3. Screening for high-risk closed loops:

[0293] • Since there may be multiple cycles in the graph, it is necessary to screen them based on risk.

[0294] • Filtering criteria: Only retain closed loops consisting of paths that contain high-priority labeled nodes or have a probability value higher than the threshold.

[0295] 4. Generate a logical closed loop: Identify and extract the high-risk cycle paths after screening, which are the logical closed loops of the high-risk conversion paths.

[0296] In one possible implementation, the correlation verification for weak nodes in the closed loop yields a complete fault evolution logic chain. The weak nodes refer to those nodes in the closed loop with high data loss rates, low monitoring sensitivity, or unstable trigger probabilities. The goal of the correlation verification is to confirm the correlation strength between weak nodes in the logical closed loop and the overall risk chain, ensuring the logic chain remains effective even with data loss or monitoring uncertainty. Feasible methods for the correlation verification are as follows:

[0297] 1. Weak node identification:

[0298] Input: The percentage of missing data calculated by S106

[0299] • Identification method: Nodes in the closed loop whose data missing ratio exceeds a threshold are selected and marked as weak nodes.

[0300] 2. Correlation verification method:

[0301] Sensitivity analysis: For weak nodes (such as those with a high proportion of missing data), the state of the node is artificially adjusted in the model (e.g., assuming it has failed or that the monitoring data is unreliable). The purpose is to examine the degree of impact of the state change of the weak node on the transformation probability or risk level of the entire closed loop. If the impact is significant, its correlation is strong and it requires close attention.

[0302] • Robustness testing: Simulates environmental impact factors and signal interference intensity (factors mentioned in S107) when the weak node fails. The purpose is to verify whether the closed-loop transformation path still holds under the most unfavorable external conditions. If the closed-loop structure can still be maintained under adverse conditions, it indicates that the closed loop has high correlation and risk.

[0303] • Logical closed-loop verification: The closed-loop structure and weak node information are input into the refined fault identification mechanism of S107. The purpose is to dynamically adjust the monitoring threshold of weak nodes using environmental factors and interference, thereby strengthening the monitoring of these key weak links in actual operation.

[0304] S106. Based on the fault evolution logic chain, simulate the impact of parameter changes on the proportion of missing data in the chain, and conduct risk assessment in combination with historical fault frequency. If the fluctuation range in the simulation continues to expand, an early warning signal is triggered to obtain the basis for risk decision-making.

[0305] By using a pre-established fault evolution model, parameter fluctuation data of key nodes in the system are acquired and preliminarily processed to obtain the basic trend of parameter fluctuation. Based on the basic trend of parameter fluctuation, and combined with the data missing ratio calculation method in the logical chain, the specific proportion of data missing is obtained to determine the degree of impact of data missing on system stability. For the specific proportion of data missing and historical fault frequency records, if the data missing ratio exceeds a preset data missing ratio threshold, a risk assessment process is triggered to determine the potential risk level. Based on the potential risk level, real-time monitoring data of fluctuation amplitude is acquired. If the fluctuation amplitude continues to increase, an early warning signal is generated to determine the abnormal state of system operation. Based on the generated early warning signal and the logical chain analysis of fault evolution, targeted decision support solutions are obtained, and the priority of countermeasures is determined. Based on the priority ranking, a random forest algorithm is used to classify the simulation analysis results, obtain risk control strategies under different scenarios, and determine the final risk mitigation path.

[0306] The parameter fluctuation data of each key node in the system obtained through the pre-established fault evolution model are specifically constructed as follows:

[0307] 1. Basic structure of the model (results from S105)

[0308] The basic structure of the model is derived from the "complete fault evolution logic chain" obtained in S105.

[0309] The fault evolution model can be a directed graph-based structure, wherein:

[0310] • Nodes: Key monitoring nodes and fault types in the power distribution system.

[0311] • Edge: High-risk transformation path between nodes (logical closed loop with completed correlation verification).

[0312] • Parameters: Each edge records the trigger probability (result of priority labeling in S105) and time interval required for the transformation (time-weighted result of S104).

[0313] 2. Model functionality expansion (adding dynamic features)

[0314] To meet the requirement of S106 “simulating parameter changes”, the model must have the ability to perform dynamic calculations and simulations.

[0315] • Model functionality / characteristics are parameter fluctuation mapping

[0316] Construction method: Map historical parameter fluctuation data and current real-time data to various key nodes in the graph.

[0317] Function: Used to "obtain parameter fluctuation data of each key node in the system".

[0318] • Model functionality / features are functions that affect the impact of missing data.

[0319] Construction method: Embed a functional relationship between the data missing ratio and the reliability of node monitoring.

[0320] Function: Used to simulate the impact of missing data ratio and calculate the degree to which missing data affects system stability.

[0321] The model's function / feature is a state transition simulation mechanism.

[0322] Construction method: Monte Carlo simulation or discrete event simulation, etc.

[0323] Function: Used to simulate how the fault state evolves along the logic chain according to the trigger probability and the time interval when a certain parameter (such as volatility) changes.

[0324] • The model's function / feature is the early warning triggering rule.

[0325] Construction method: Embed the warning thresholds mentioned in S102 and S106 (such as preset fluctuation range thresholds).

[0326] Function: Used to determine whether "if the fluctuation range continues to expand during the simulation process, an early warning signal will be triggered".

[0327] 3. Summary

[0328] Therefore, the construction of the pre-established fault evolution model is to use the complete fault evolution logic chain obtained in S105 as the skeleton, and then embed historical fluctuation data, data missing impact function and dynamic state transition simulation mechanism to make it a calculation tool that can perform real-time risk assessment and early warning signal triggering for high-risk transformation paths.

[0329] 4. In one possible implementation, the step of obtaining the specific proportion of missing data based on the basic trend of parameter fluctuations and the data missing ratio calculation method in the logic chain, and determining the degree of impact of data missing data on system stability, typically uses a simple statistical indicator to quantify the difference between the expected data points and the actual data points collected within a certain time window. The specific calculation method is as follows:

[0330] A. Calculation method for a single key node

[0331] For a single critical node (corresponding to a certain sensor or monitoring point) in the fault evolution model within a certain time window T:

[0332]

[0333] The total number of data points to be collected depends on the system's preset sampling frequency and the length of the time window T.

[0334] B. Calculation method for the entire logic chain

[0335] For the complete fault evolution logic chain (containing multiple key nodes):

[0336] It can calculate the average percentage of missing data for all key nodes within a time window T.

[0337] A more advanced calculation method is to use a weighted average: assign higher weights to key nodes on high-risk transformation paths (determined by S104) to highlight the risk of missing data.

[0338] The purpose of describing the impact of missing data on system stability is to convert the quantitative value (proportional value) of missing data into a quantitative value of the credibility or risk of the fault evolution logic chain.

[0339] A. The specific meaning of "degree of effect" may refer to one or a combination of the following:

[0340] 1. Decrease in the credibility of the logical chain: The higher the proportion of missing data, the greater the decrease in the credibility of the logical chain.

[0341] 2. Increase in risk assessment value: The higher the proportion of missing data, the greater the uncertainty of the system's risk assessment value or failure probability value.

[0342] B. Feasible methods for determining the extent of effect

[0343] Determining the degree of effect usually requires pre-establishing a mapping function or lookup table.

[0344] 1. Mapping function method

[0345] Pre-set a function: For example, using a non-linear function, the impact (such as risk uncertainty) increases exponentially once the missing data ratio exceeds a certain threshold. This smoothly translates the missing data ratio into an effect on system stability.

[0346] 2. Risk Impact Analysis

[0347] The data missing percentage is used as a conditional input to the fault evolution model (Step 1, S106). Step 3, S106: If the missing percentage exceeds a threshold, a risk assessment is triggered. During simulation, the model translates a higher data missing percentage into a more unstable range of parameter fluctuations, thereby amplifying the possibility of a continuously expanding range of fluctuations in the simulation, ultimately leading to a higher risk assessment result.

[0348] 3. Correlation verification feedback

[0349] Step S105, step seven, verifies the correlation of weak nodes. If a node with missing data is determined by S105 to be a weak node with extremely strong correlation, its impact level is automatically set to the highest level. This ensures that data loss in the most critical weak links of the logical chain is given the highest risk priority.

[0350] In summary, the objective of S106 is to simulate the impact of missing data on risk. Therefore, the degree of impact refers to the quantified value of the negative impact of missing data on the reliability of transformation paths and risk uncertainty in the fault evolution model.

[0351] In one possible implementation, regarding the specific proportion of missing data and historical fault frequency records, if the data missing proportion exceeds a preset data missing proportion threshold, a risk assessment process is triggered to determine the potential risk level. The preset data missing proportion threshold is set as follows:

[0352] A. Setting the initial threshold (based on reliability requirements)

[0353] Method: System reliability requirements method

[0354] • Steps: Based on the minimum data quality requirements of the power distribution system, set a maximum tolerable percentage of missing data. For example, if the system requires monitoring data to be available 95% of the time, then the tolerable percentage of missing data is 5%.

[0355] • Initial threshold: Usually set based on engineering experience, for example, 3%-5% is a reasonable initial value.

[0356] • Threshold < 3%: Considered high data quality, usually does not trigger risk assessment.

[0357] • Threshold > 5%: This is considered low data quality and requires triggering a risk assessment process.

[0358] • Purpose: To establish a baseline and ensure that risk processes can be initiated immediately should data loss affect the core reliability of system monitoring.

[0359] B. Optimization of dynamic thresholds (based on criticality and risk)

[0360] This threshold should not be fixed, but should be dynamically adjusted based on the location of missing data and associated risks.

[0361] • Optimization criteria such as logical chain criticality:

[0362] If data loss occurs at high-priority critical nodes or weak nodes identified in S105, adjust the threshold by lowering it. Even if the loss rate is only 1% or 2%, a risk assessment should be triggered immediately, as data loss at these nodes has the greatest impact on the entire logical chain.

[0363] Optimization criteria include, for example, historical failure frequency.

[0364] Combine historical failure frequency records. If a device has a slightly higher missing rate but also a high historical failure frequency, adjust: tighten (lower) the threshold. The combination of missing data and high-risk historical records should be considered a more serious risk signal.

[0365] • Optimization criteria such as environmental impact factors

[0366] Step S107: Adjust monitoring thresholds for environmental impact factors. Adjustment: When severe environmental conditions (such as strong storms or high signal interference) cause unstable data transmission, the thresholds can be appropriately relaxed (raised) to avoid excessive warnings due to environmental factors.

[0367] If the data missing ratio exceeds the preset data missing ratio threshold, a risk assessment process is triggered. The specific method for determining the potential risk level is as follows:

[0368] The determination of the potential risk level is the core output of S106, ultimately used to generate the basis for risk decision-making. This level is a comprehensive assessment used to quantify the likelihood and severity of the failure evolution under the current data gap condition.

[0369] A. Risk assessment process and methods

[0370] When the proportion of missing data exceeds the threshold, the triggered risk assessment process should primarily rely on the pre-established fault evolution model (Step 1, S106) for simulation.

[0371] 1. Input data: data missing rate, historical fault frequency records, and current parameter fluctuation trend.

[0372] 2. Simulated State Uncertainty: When simulating state transitions in the model, the proportion of missing data is used as an uncertainty factor for the state transition probability or parameter fluctuation range. The higher the proportion of missing data, the larger the fluctuation range simulated by the model, and the more uncertain the probability of state transitions.

[0373] 3. Evaluation of simulation results:

[0374] • Simulation: Run a fault evolution model to simulate faults under conditions of high data loss (such as Monte Carlo simulation).

[0375] • Triggering warning signal count: The number or probability of triggering a warning signal when the "fluctuation range continues to expand" in the statistical simulation.

[0376] B. Specific indicators of potential risk level (quantification of the degree of impact)

[0377] The potential risk level is typically a graded indicator (such as Level I, Level II, Level III, or Low, Medium, High). It is a quantification of simulation results. Specific risk level indicators include:

[0378] • Likelihood of occurrence: The probability that a potential risk will be successfully transformed under the current condition of missing data.

[0379] Judgment basis: The increase in the cumulative probability value of the high-risk transformation path (determined in S104) in the model simulation.

[0380] • Severity of impact: The severity of the consequences on the system if the potential risk is successfully converted.

[0381] Judgment criteria: Combining historical fault frequency records (such as power outage time and equipment damage extent) and hidden feature markers (risks with high concealment are judged to be more serious).

[0382] • Risk level: A classification based on a combination of likelihood and severity.

[0383] Judgment criteria: Low: Low probability, low severity. Medium: High probability or high severity. High: High probability and high severity. Very high: High proportion of missing data and continuously expanding fluctuation range in simulation, high probability of triggering warning signal.

[0384] S107. Based on risk decision-making, update the power distribution system monitoring rules, adjust the monitoring thresholds according to environmental impact factors and signal interference intensity, and optimize the monitoring logic by combining hidden feature markings to obtain a refined fault identification mechanism.

[0385] By collecting environmental impact factors and signal interference intensity, real-time environmental and interference data of the power distribution system are obtained, identifying key variables affecting monitoring effectiveness. If any variable exceeds a preset safety threshold, it is marked as an anomalous variable, resulting in a set of anomalous variables. For this set, relevant data on hidden feature markers are acquired, and the potential correlation between anomalous variables and fault types is determined based on the application of these markers, outputting the correlation analysis results. From the correlation analysis results, high-risk variables related to the fault identification mechanism are extracted, and a support vector machine algorithm is used to classify these high-risk variables, obtaining the classified fault risk level. Based on the classified fault risk level and the principle of monitoring logic optimization, the rule base for power distribution system monitoring is updated, determining a new combination of monitoring rules. Real-time operating data is obtained and matched using this new combination of rules. If the matching result shows an anomaly, the fault identification mechanism is triggered, outputting a specific fault type judgment. Based on the fault type judgment result, the monitoring threshold is dynamically adjusted, and an optimized monitoring threshold configuration is obtained using a dynamic threshold update strategy.

[0386] In one possible implementation, by collecting environmental impact factors and signal interference intensity to obtain real-time environmental and interference data during the operation of the power distribution system, and identifying key variables affecting the monitoring effect, the key variables refer to environmental or interference factors that have a significant impact on the accuracy of fault early warning (i.e., fault identification rate, false alarm rate, and missed alarm rate). Determining these variables requires using statistical analysis and machine learning methods to correlate them with the system's monitoring performance.

[0387] 1. Initial Data Preparation

[0388] First, all collected environmental and interference data needs to be correlated with the system's monitoring performance indicators:

[0389] • Independent variable (X): All collected environmental impact factors (e.g., temperature, humidity, salt spray, wind speed) and signal interference intensity (e.g., electromagnetic interference, noise power).

[0390] • Dependent variable (Y): The system's monitoring performance indicators, for example:

[0391] · False alarm rate (the proportion of normal fluctuations that are mistakenly identified as abnormal within a certain time window).

[0392] · : False negative rate (the proportion of actual faults that are not identified within a certain time window).

[0393] · Fault identification accuracy.

[0394] 2. Methods for determining key variables

[0395] A. Statistical correlation analysis (quantifying the intensity of influence)

[0396] • Method: Calculate the statistical correlation coefficient (e.g., Pearson correlation coefficient) between each environmental / disturbance variable (X) and the monitoring effect indicator (Y).

[0397] Judgment criteria:

[0398] • High correlation screening: If the absolute value of the correlation coefficient between a variable and the false positive rate or false negative rate exceeds a preset correlation threshold (e.g., 0.7).

[0399] • Example: If a strong positive correlation is found between humidity and the false alarm rate of insulation faults, then "humidity" is identified as a key variable.

[0400] B. Machine Learning Feature Importance Method (Eliminating Redundant Variables)

[0401] • Model building: Use machine learning models such as random forests or gradient boosting trees to train the model to predict the false positive or false negative rate of the system.

[0402] • Feature Score: The model calculates a feature importance score for each input variable (environment / interference factor).

[0403] • Judgment criteria: The variable with the highest feature importance score is identified as the key variable. This excludes variables that, although collected, have little impact on the accuracy of system monitoring.

[0404] C. Risk correlation verification method (combining the results of S103)

[0405] • Method: Combine the potential impact weights of the environment on the fault type determined in step S103.

[0406] • Judgment criteria: If an environmental factor (such as high temperature) is identified by S103 as having a high weighting on the failure type on the high-risk transformation path, then even if the factor is not highly correlated with the real-time false alarm rate, it should be identified as a key variable and used for dynamic adjustment in S107 because of its potential risk contribution.

[0407] 3. Subsequent Applications

[0408] Once the key variables are identified, the system will proceed to the next step:

[0409] 1. Quantization mapping: Establish the functional relationship between the real-time values ​​of these key variables and the adjustment amount of the monitoring threshold.

[0410] 2. Dynamic adjustment: Based on the real-time monitoring values ​​of key variables, the monitoring thresholds used in steps S101, S102, etc., are dynamically adjusted to form a "refined fault identification mechanism".

[0411] In one possible implementation, if any variable exceeds a preset safety range threshold, it is marked as an abnormal variable, thus obtaining a set of abnormal variables. The preset safety range threshold is set as follows:

[0412] 1. Setting the initial threshold (based on specifications and historical data)

[0413] Method 1: Equipment / Environment Specification Method (Static Threshold)

[0414] • Steps: Consult the design specifications for power distribution system equipment (such as sensors and lines) and the environment.

[0415] • Initial threshold: Use the highest / lowest safe operating limits specified in these standards as the threshold.

[0416] For example: the upper limit of safe temperature in the transformer room (e.g., 50°C), and the upper limit of safe humidity in the line environment (e.g., 95%).

[0417] • Function: Ensure that environmental / interference data is immediately flagged as abnormal once it exceeds the device's design tolerance.

[0418] Method 2: Historical Monitoring Effect Method (Statistical Threshold)

[0419] • Step: Analyze the values ​​of environmental or interference variables (X variables) when the historical monitoring accuracy (Y variable in Step 1 of S107) began to decline significantly.

[0420] • Initial threshold: Set to a value slightly lower than the value at which the "monitoring effect begins to deteriorate".

[0421] • Purpose: To ensure that environmental / disturbing variables are marked as anomalous when the monitoring effect is about to be affected.

[0422] 2. Optimization of dynamic thresholds (based on risk and correlation)

[0423] Once the key variables in step one of S107 are determined, their safety range thresholds must be dynamically adjusted:

[0424] Optimization criteria such as the correlation between fault types

[0425] In conjunction with the potential impact weights of S103: If an environmental factor (such as salt spray) has an extremely high impact weight on high-risk failure types, the threshold should be tightened (more stringent). Even if the salt spray concentration has not yet reached the upper limit for equipment damage, it should be marked as abnormal as long as it reaches a concentration that could trigger a high-risk failure.

[0426] Optimization criteria include monitoring system reliability.

[0427] Based on the data missing rate in S106: the data missing rate increases significantly under specific environmental conditions (e.g., low temperature). Adjustment: Tighten the threshold. Raise the safe range threshold for this environmental factor to allow for earlier intervention (e.g., heating) to ensure the reliability of data acquisition.

[0428] • Optimization is based on feedback from mechanisms such as refining.

[0429] In conjunction with the threshold adjustment feedback in S107: If a critical variable frequently exceeds its threshold value, causing the system to frequently adjust the monitoring threshold, but the improvement in fault identification is not significant, the solution may be to reassess the threshold for that variable or relax the threshold to reduce invalid "abnormal variable" flags.

[0430] Through this dynamic adjustment, the threshold can ensure that the "abnormal variables" that the system focuses on are the key variables that have the greatest impact on the effectiveness of early warning monitoring.

[0431] In one possible implementation, the step of obtaining real-time operational data through a new combination of monitoring rules for matching, and triggering a fault identification mechanism to output a specific fault type judgment if the matching result shows an anomaly, involves comparing the real-time data with optimized monitoring thresholds and rules to determine whether an anomaly has occurred. The specific matching method is as follows:

[0432] 1. Input data: Real-time operating data of the power distribution system (current, voltage, temperature, etc.).

[0433] 2. New monitoring rule combination: S107 Steps three and four, through dynamic adjustment of the optimized monitoring value configuration. This includes:

[0434] • Dynamic thresholds: Abnormal fluctuation thresholds (S101) and abnormal fluctuation frequency thresholds (S102) adjusted for key variables (such as high temperature and strong interference).

[0435] • Covert Feature Rules: Optimized covert feature recognition rules (S104 markers) are used to identify slow or multivariate coupling anomalies that are not easily detected.

[0436] 3. Matching process:

[0437] • Multiple condition comparison: Real-time data will be compared sequentially through this series of dynamic thresholds and optimization rules.

[0438] • Matching result: If the real-time data violates any of the optimized rules (for example, the real-time current fluctuation frequency exceeds the dynamically adjusted S102 threshold, or the real-time parameters meet the rules of the hidden feature marker), the matching result will show an anomaly and trigger the subsequent fault identification mechanism.

[0439] If the matching result shows an anomaly, the fault identification mechanism, which is the core output of the entire method, is triggered. It is responsible for determining the specific type of fault based on the matched anomaly. This fault identification mechanism can be based on reverse reasoning using a fault evolution model and logical chain. Specifically:

[0440] 1. Input: Matched abnormal data and complete fault evolution logic chain (output of S105).

[0441] 2. Reverse reasoning: The system compares the feature vectors (such as fluctuation frequency and amplitude) of the matched abnormal data with the key nodes and transformation paths on the fault logic chain.

[0442] 3. Determine the type of fault:

[0443] • If the abnormal data meets the “triggered” condition of a certain node in the logic chain, then the fault type corresponding to that node (such as “overload” or “insulation aging”) is the preliminary judgment result.

[0444] • If the abnormal data shows hidden characteristics, the corresponding hidden fault type is directly output through the hidden characteristic marker (S104).

[0445] 4. Output: Specific fault type determination.

[0446] The fault identification mechanism outputs a specific fault type judgment. The result of the fault type judgment comes from the fault type association dataset (established in S103), which contains the names or codes of various faults that may occur in the power distribution system. The fault type judgment result specifically includes:

[0447] Physical faults: short circuit, grounding, open circuit, overload, equipment damage, etc.

[0448] • Characteristic degradation faults: insulation aging, loose joints, increased contact resistance.

[0449] • Environmentally induced faults: flashover, pollution flashover (induced by environmental factors such as salt spray and humidity).

[0450] • Covert faults: Small-amplitude, slow-evolving faults (identified by covert feature markers).

[0451] In one possible implementation, dynamically adjusting the monitoring threshold based on the fault type determination means that different fault types correspond to different monitoring threshold adjustments, reflecting the refinement and targeting of the threshold adjustment. The specific adjustment logic is as follows:

[0452] • Objective: To make monitoring thresholds more focused on current or impending high-risk fault types.

[0453] Example:

[0454] 1. If the judgment result is "insulation aging": The system will tighten the environmental thresholds related to temperature and humidity, and at the same time tighten the thresholds for small and slow fluctuations in current and voltage, in order to improve the sensitivity to the trend of insulation deterioration.

[0455] 2. If the judgment result is "loose connection": The system will focus on tightening the thresholds related to vibration and sudden temperature rise, and give priority to high-frequency, random signal interference (because loose connection may lead to poor contact and electric arc).

[0456] 3. If the judgment result is "overload": The system will focus on tightening the threshold of current amplitude, but may relax the tolerance for small temperature fluctuations.

[0457] In this way, the system avoids using a fixed set of thresholds and instead dynamically optimizes monitoring resources and sensitivity based on the specific fault threats it faces.

[0458] Traditional fault warning systems often use a fixed, universal monitoring threshold (e.g., an alarm is triggered when current fluctuations exceed 5%). However, this application, in step S107, optimizes monitoring sensitivity based on specific risks. Specifically:

[0459] 1. The logic of the adjustment

[0460] When the fault identification mechanism outputs a specific fault type (e.g., "insulation aging"), the system will adjust the associated monitoring thresholds accordingly based on this information.

[0461] Fault type determination result: Insulation aging

[0462] • Example of adjustment strategy: Tighten the thresholds associated with thermal effects and slow changes.

[0463] • Adjusted monitoring thresholds: slow fluctuation threshold for current / voltage (tightened); abnormal fluctuation threshold for temperature (tightened).

[0464] Fault type determination result: Loose connection

[0465] • Example of adjustment strategy: Tighten the thresholds related to transient and random disturbances.

[0466] • Adjusted monitoring thresholds: High-frequency signal interference intensity threshold (tightened); High-frequency abnormal fluctuation frequency threshold (tightened).

[0467] • Fault type assessment result: Overload risk

[0468] • Example of adjustment strategy: Tighten the threshold related to current amplitude, and loosen the threshold related to slow changes.

[0469] • Adjusted monitoring thresholds: Abnormal current fluctuation threshold (tightened); slow temperature fluctuation tolerance (relaxed).

[0470] • Fault type determination result: External flashover / Pollution safety

[0471] • Example of adjustment strategy: Tighten the thresholds related to environmental factors (humidity, salt spray).

[0472] • Adjusted monitoring thresholds: Environmental impact factor safety range threshold (tightened); voltage fluctuation threshold (tightened).

[0473] 2. Advantages of dynamic adjustment

[0474] • Improve the targeting of early warnings: Focus resources (monitoring sensitivity) on the most likely faults to occur, thereby improving the efficiency of risk identification.

[0475] • Reduce false alarm rate: Only minor fluctuations that are not consistent with the current high-risk fault types will have their thresholds relaxed, thereby reducing false alarms for non-critical events.

[0476] • Achieve refined monitoring: Ensure that the key links (S105) on the high-risk conversion path are given the highest monitoring priority and the strictest threshold control.

[0477] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the concept of this application. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for power distribution line fault pre-warning and analysis, characterized in that, The method includes: By extracting parameter change sequences from electrical parameters in the power distribution system in real time, performing feature decomposition on the frequency and amplitude range of abnormal fluctuations, analyzing signal interference intensity, and obtaining a preliminary distribution of abnormal patterns; Based on the preliminary distribution of abnormal patterns, combined with the duration and time window distribution, the parameter change trends are classified. If the frequency of abnormal fluctuations in the classification results exceeds the preset threshold, the potential chain reaction initiation point is determined. By identifying potential chain reaction starting points, we obtain the historical failure frequency and environmental impact factors corresponding to the associated equipment identifiers, construct a failure type association dataset, analyze the failure type hierarchy, and obtain the type transformation direction sequence. Based on the type conversion direction sequence, the conversion time interval and path weight allocation are integrated to calculate the probability of each conversion path. If a path corresponds to a hidden feature label and the probability is higher than a preset threshold, then a high-risk conversion path is determined. By identifying high-risk transformation paths, a fault evolution graph is generated. Based on node connection density and node influence radius, graph structure analysis technology is used to sort out path cycle patterns and determine the logical chain of fault evolution. Based on the fault evolution logic chain, the impact of parameter changes on the proportion of missing data in the chain is simulated, and risk assessment is carried out in combination with historical fault frequency. If the fluctuation range in the simulation continues to expand, an early warning signal is triggered, and the basis for risk decision-making is obtained. By updating the power distribution system monitoring rules based on risk decision-making, adjusting the monitoring thresholds according to environmental impact factors and signal interference intensity, and optimizing the monitoring logic by combining hidden feature markings, a refined fault identification mechanism is obtained.

2. The method for power distribution line fault pre-warning and analysis according to claim 1, characterized in that, The process involves extracting parameter change sequences from electrical parameters in the power distribution system through real-time monitoring, performing feature decomposition on the frequency and amplitude range of abnormal fluctuations, analyzing signal interference intensity, and obtaining a preliminary distribution of abnormal patterns, including: The current and voltage signals are collected in real time by sensors in the power distribution system. The collected raw data is preprocessed using digital processing technology to obtain a preliminary set of electrical parameter data. Extract the change sequence data from the initially organized electrical parameter dataset, identify the abnormal fluctuation parts in the change sequence, and use a preset threshold range for preliminary screening to determine the set of potential abnormal points; For the selected set of anomaly points, the frequency and amplitude range of the abnormal fluctuations are analyzed, and the anomaly points are decomposed in multiple dimensions using feature decomposition technology to obtain the decomposed feature vector set. By combining the decomposed feature vector group with time-series processing technology, the intensity of signal interference is quantitatively analyzed. If the signal interference intensity exceeds the preset signal interference intensity threshold, it is determined to be a high-risk interference point, and a list of high-risk interference points is obtained. Based on the list of high-risk interference points, the correspondence between signal interference and abnormal patterns is analyzed. The support vector machine algorithm is used to classify the abnormal patterns and determine the preliminary distribution categories of the abnormal patterns. Based on the preliminary distribution categories of abnormal patterns, corresponding abnormal pattern distribution maps are generated. The correlation between abnormal fluctuations and signal interference is visualized and mapped through the map data to obtain the distribution characteristics of abnormal patterns. By comparing and analyzing the distribution characteristics of abnormal patterns with historical time-series data, if there is a significant deviation between the distribution characteristics and historical abnormal patterns, they are marked as new abnormal patterns, and the potential risk level of the new abnormal patterns is determined.

3. The method of claim 1, wherein, The parameter change trend is classified based on the preliminary distribution of abnormal patterns, combined with the duration and time window distribution. If the frequency of abnormal fluctuations in the classification results exceeds a preset threshold, the potential chain reaction initiation point is determined, including: Obtain abnormal pattern data recorded in the system, organize preliminary distribution characteristics, and use data grouping technology to preliminarily divide the abnormal patterns according to time windows to obtain the grouped abnormal dataset. Extract relevant information on duration and time window from the grouped abnormal datasets, analyze the parameter changes in each group one by one, and determine the trend characteristics of change in each group; Based on the characteristics of the changing trends, the data in each group are classified and processed to generate classification result data. The frequency of abnormal fluctuations in the classification results is statistically analyzed to obtain the fluctuation frequency value. Obtain the fluctuation frequency value. If the fluctuation frequency value exceeds the preset abnormal fluctuation frequency threshold, it is determined that there is a potential chain reaction starting point within the group and it is marked as a high-risk group. Extract abnormal patterns and parameter change information from the data of high-risk groups, and use the support vector machine algorithm to conduct further pattern analysis on the data within the high-risk groups to determine the possible trigger points of the chain reaction; Based on the possible trigger points of the chain reaction, corresponding time windows and monitoring strategies within the duration are generated, and data related to the trigger points are tracked in real time to obtain dynamic change data of abnormal fluctuations. By dynamically changing data, the calculation results of the fluctuation frequency are continuously updated. If the updated fluctuation frequency still exceeds the preset abnormal fluctuation frequency threshold, the exact location of the chain reaction starting point is confirmed, and corresponding early warning information is generated.

4. The method of claim 1, wherein, The process involves obtaining historical fault frequencies and environmental impact factors corresponding to associated equipment identifiers through potential chain reaction starting points, constructing a fault type association dataset, analyzing the fault type hierarchy, and obtaining a type transformation direction sequence, including: By extracting data from the equipment identification database, historical fault records and fault frequency information related to the initiation point of the chain reaction are obtained, and a preliminary fault distribution pattern is determined. Based on the preliminary failure distribution pattern and combined with environmental impact factor data, the correlation between environmental impact and failure frequency is analyzed to obtain the potential impact weight of the environment on failure type. Using Bayesian inference, an associated dataset is constructed based on fault type and environmental influence weights to derive the hierarchical structure of fault types; By analyzing the hierarchical structure of fault types, we can determine the transformation direction between types, obtain the probability distribution of the transformation direction, and identify the main transformation path. Based on the results of the conversion path and combined with historical fault data, if the conversion probability of a certain fault type is higher than the preset conversion probability threshold, it is marked as a high-risk conversion direction and a risk warning message is output. Acquire data on high-risk conversion directions, analyze their correspondence with equipment identifiers, generate equipment-level fault prevention strategies, and obtain targeted prevention and control measures; By analyzing the implementation effectiveness data of prevention and control measures, we update the fault type hierarchy and transformation direction sequence in the associated dataset, and continuously optimize the accuracy of subsequent analyses.

5. The method of claim 1, wherein, The step involves calculating the probability of each transformation path based on the type transformation direction sequence, the transformation time interval, and the path weight allocation. If a path corresponds to a hidden feature marker and its probability is higher than a preset threshold, then a high-risk transformation path is identified, including: Obtain data records of conversion type and sequence direction, extract sequence information of each conversion path and classify and organize them to obtain a preliminary path classification set; Based on the initial path classification set, the corresponding interval time data is obtained. Combining the path weight and weight allocation rules, the interval time of each path is weighted to obtain the time-weighted result of each path. The probability value of each conversion path is calculated based on the time-weighted results. A pre-established probability model is used to assess the probability and determine the probability value of each path. For the probability value records, obtain the hidden features and feature label information of each path. If the hidden features of a path match the feature label, then mark it as a potential risk path. Based on the probability values ​​of potential risk paths, if the probability value of a path exceeds a preset probability threshold, it is determined to be a high-risk path, and a list of high-risk paths is output. By listing high-risk paths, detailed data on relevant conversion types and sequence directions are obtained, path risk correlation analysis records are generated, and the final risk path distribution information is determined.

6. The method of claim 1, wherein, The process involves generating a fault evolution graph through high-risk transformation paths. Based on node connection density and node influence radius, graph structure analysis techniques are used to analyze path loop patterns and determine the logical chain of fault evolution, including: By using a pre-established fault data archive, relevant records of high-risk conversion paths are obtained, and key nodes and conversion events in the records are classified and labeled to obtain a preliminary fault evolution map framework. Based on the preliminary fault evolution map framework, graph structure analysis technology is used to perform depth-first traversal of the connection relationships between nodes to determine the distribution pattern of node connection density. By analyzing the distribution pattern of node connection density, the role weight of each node in the fault evolution graph is analyzed. By using a weighted calculation method, the triggering conditions of key nodes are determined, and the initial sequence of the fault logic chain is obtained. For the initial sequence of the fault logic chain, the feature data of the path loop pattern is obtained. If there are repeated loop paths in the feature data, the paths are merged to determine the optimized loop pattern recognition result. Based on the optimized loop pattern recognition results, potential risk points in high-risk transformation paths are identified, and the correlation strength between risk points and fault logic chains is determined by traversal comparison, thus obtaining the structural framework of the evolution chain. By analyzing the structural framework of the evolution chain, the key links in the transformation path are analyzed. If the trigger probability of a key link is higher than the preset trigger probability threshold, it is prioritized and the final fault evolution map is determined. Based on the final fault evolution map, a logical closed loop of high-risk transformation path is generated. The weak nodes in the closed loop are verified for correlation to obtain a complete fault evolution logical chain.

7. The method of claim 1, wherein the method further comprises: The simulation, based on the fault evolution logic chain, simulates the impact of parameter changes on the proportion of missing data in the chain. Risk assessment is then conducted by combining this with historical fault frequencies. If the fluctuation range in the simulation continues to expand, an early warning signal is triggered, providing a basis for risk decision-making, including: By using a pre-established fault evolution model, parameter fluctuation data of each key node in the system are obtained and preliminarily sorted out to obtain the basic trend of parameter fluctuation. Based on the basic trend of parameter fluctuations and combined with the method for calculating the proportion of missing data in the logical chain, the specific proportion of missing data is obtained to determine the degree of impact of missing data on system stability. For specific data missing percentages and historical failure frequency records, if the data missing percentage exceeds the preset data missing percentage threshold, a risk assessment process is triggered to determine the potential risk level. Based on the potential risk level, real-time monitoring data of the fluctuation amplitude is obtained. If the fluctuation amplitude continues to increase, an early warning signal is generated to determine the abnormal state of the system operation. By analyzing the generation results of early warning signals and the logical chain of fault evolution, targeted decision support solutions can be obtained, and the priority of countermeasures can be determined. Based on the priority ranking, the simulation analysis results are classified using the random forest algorithm to obtain risk control strategies under different scenarios and determine the final risk mitigation path.

8. The method for early warning and analysis of power distribution line faults according to claim 1, characterized in that, The aforementioned mechanism updates the power distribution system monitoring rules based on risk decision-making, adjusts monitoring thresholds according to environmental impact factors and signal interference intensity, and optimizes monitoring logic by combining hidden feature marking, resulting in a refined fault identification mechanism, including: By collecting environmental impact factors and signal interference intensity, real-time environmental and interference data during the operation of the power distribution system are obtained, and key variables affecting the monitoring effect are identified. If any variable exceeds a preset safety threshold, it is marked as an abnormal variable, thus obtaining a set of abnormal variables; For the set of anomalous variables, obtain relevant data on hidden feature markers, and determine the potential association between anomalous variables and fault types by combining the application of feature markers, and output the association analysis results; From the correlation analysis results, high-risk variables related to the fault identification mechanism are extracted, and the support vector machine algorithm is used to classify the high-risk variables to obtain the classified fault risk level. Based on the classified fault risk levels and in accordance with the principle of monitoring logic optimization, the rule base for power distribution system monitoring is updated, and new combinations of monitoring rules are determined. By combining new monitoring rules, real-time operating data is obtained and matched. If the matching result shows an anomaly, the fault identification mechanism is triggered, and the specific fault type is output. Based on the fault type determination, the monitoring threshold is dynamically adjusted, and combined with the dynamic threshold update strategy, an optimized monitoring threshold configuration is obtained.

Images