Monitoring information analysis method and system of power distribution terminal

Abstract

The invention discloses a power distribution terminal monitoring information analysis method and system, and the method comprises the steps: obtaining a monitoring data flow in an initial time period, and constructing a normal data feature library associated with a multi-dimensional working condition; acquiring and analyzing a current monitoring data stream in real time, and separating abnormal and normal data features based on a dynamic abnormal judgment threshold; when an unknown anomaly is found, searching associated normal data features in front and back time windows of the unknown anomaly, and calculating the weighted matching degree of the associated normal data features and the reference fragment in the feature library; and if the matching degree exceeds a preset threshold value, taking the associated normal data features as a new reference, adding the new reference into the feature library, and updating and transmitting. According to the method, working condition self-adaptive precise monitoring and abnormal intelligent identification of the operation state of the power distribution terminal are realized, the system can continuously learn a legal new operation mode through a dynamic self-evolution mechanism of the knowledge base, the technical problems that the false alarm rate of a fixed threshold value is high and the system cannot adapt to complex working condition changes are effectively solved, and the system is suitable for large-scale popularization and application. And the state evaluation accuracy and the operation and maintenance intelligent level are obviously improved.

Description

Technical Field

[0001] This invention belongs to the field of power distribution terminal monitoring and management technology, and particularly relates to a method and system for analyzing monitoring information of power distribution terminals. Background Technology

[0002] With the development of smart distribution networks, the number of distribution terminals has surged, and the monitoring parameters they collect, such as voltage, current, temperature, and switch status, form a continuous, high-dimensional time-series data stream. Timely and accurate identification of information characterizing equipment anomalies or grid disturbances from these data streams is crucial for ensuring power supply safety and reliability.

[0003] Currently, the analysis of monitoring data from power distribution terminals largely relies on threshold alarms or simple trend analysis. Threshold alarm methods require setting fixed upper and lower safety limits for each parameter. However, the normal operating status of power distribution equipment changes with various conditions such as load, environment, and operating mode. Fixed thresholds are difficult to adapt to this dynamism, easily leading to false alarms or missed alarms. Simple trend analysis, on the other hand, lacks in-depth analysis of the coupling relationships between multiple parameters and the complete evolution process of abnormal events (occurrence, development, and recovery).

[0004] Another approach is to use machine learning models for anomaly detection. However, these methods typically require a large number of labeled anomaly samples for training. In real-world power grids, anomaly samples are scarce and of unknown types, leading to difficulties in model training and insufficient generalization ability. Furthermore, existing methods often lack effective knowledge update mechanisms after discovering unknown types of anomalies or when the system enters a new steady-state operating mode, making it difficult to achieve self-learning and continuous optimization of the monitoring system.

[0005] Therefore, there is an urgent need for an intelligent analysis method for power distribution terminal monitoring information that can adapt to different operating conditions, accurately identify known and unknown anomalies, and have self-updating capabilities. Summary of the Invention

[0006] To address the shortcomings in the aforementioned background technology, this invention provides a method and system for analyzing monitoring information of power distribution terminals, aiming to achieve adaptive analysis of monitoring data based on operating conditions, accurate identification of anomalies, and dynamic evolution of the feature database.

[0007] In a first aspect, the present invention provides a method for analyzing monitoring information of a power distribution terminal, comprising:

[0008] Acquire the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period;

[0009] The first monitoring data stream is parsed to construct a normal data feature library, which contains multiple baseline data feature fragments of each monitoring parameter under no abnormal conditions.

[0010] Acquire the second monitoring data stream of the power distribution terminal in the current time period, parse the second monitoring data stream to obtain the abnormal data feature set and the current normal data feature set;

[0011] Determine whether there exists a benchmark data feature fragment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set;

[0012] If it does not exist, then search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library.

[0013] Determine whether the matching degree is greater than a preset threshold;

[0014] If the value exceeds a preset threshold, the data segment corresponding to the at least one current normal data feature is added to the normal data feature library, and the updated normal data feature library and the abnormal data features associated with the at least one current normal data feature are transmitted to the user terminal.

[0015] Secondly, the present invention provides a monitoring information analysis system for a power distribution terminal, comprising:

[0016] The acquisition module is configured to acquire the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period;

[0017] The first parsing module is configured to parse the first monitoring data stream and construct a normal data feature library, wherein the normal data feature library contains multiple baseline data feature fragments of each monitoring parameter under no abnormal conditions.

[0018] The second parsing module is configured to acquire the second monitoring data stream of the power distribution terminal in the current time period, parse the second monitoring data stream, and obtain the abnormal data feature set and the current normal data feature set;

[0019] The first judgment module is configured to determine whether there is a benchmark data feature fragment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set;

[0020] The calculation module is configured to, if it does not exist, search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library.

[0021] The second judgment module is configured to determine whether the matching degree is greater than a preset threshold.

[0022] The update module is configured to add the data segment corresponding to the at least one current normal data feature to the normal data feature library if the value is greater than a preset threshold, and transmit the updated normal data feature library and the abnormal data features associated with the at least one current normal data feature to the user terminal.

[0023] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the monitoring information analysis method for a power distribution terminal according to any embodiment of the present invention.

[0024] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the power distribution terminal monitoring information analysis method according to any embodiment of the present invention.

[0025] The monitoring information analysis method and system for distribution terminals in this application achieves accurate and adaptive monitoring of the operating status of distribution terminals by constructing a normal data feature library dynamically associated with multi-dimensional operating conditions and employing weighted similarity matching and dynamic anomaly judgment thresholds. Furthermore, by deeply analyzing the matching degree between normal data features associated with unknown anomaly events and historical benchmarks, the method intelligently decides whether to include feature segments representing new steady states into the feature library, forming a complete technical closed loop of data parsing, anomaly identification, association analysis, and knowledge self-evolution. This method effectively solves the industry problems of high false alarm rates and inability to adapt to complex and changing operating conditions caused by traditional fixed threshold alarms, significantly improving the accuracy of state assessment and context awareness. The dynamic update mechanism of the knowledge base enables the system to continuously learn and evolve, automatically tracking and learning legitimate new states of the distribution system generated by equipment aging, network reconstruction, or operational mode optimization. This achieves a leap from a "static rule base" to a "dynamic intelligent agent," thereby significantly reducing operation and maintenance costs, improving the timeliness and reliability of fault warnings, and providing core technical support for building a highly reliable intelligent distribution network with predictive maintenance capabilities. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1A flowchart illustrating a method for analyzing monitoring information of a power distribution terminal according to an embodiment of the present invention;

[0028] Figure 2 This is a structural block diagram of a power distribution terminal monitoring information analysis system provided in an embodiment of the present invention;

[0029] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figure 1 The diagram shows a flowchart of a monitoring information analysis method for a power distribution terminal according to this application.

[0032] like Figure 1 As shown, the method for analyzing monitoring information from power distribution terminals specifically includes the following steps:

[0033] Step S101: Obtain the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period.

[0034] Step S102: parse the first monitoring data stream and construct a normal data feature library. The normal data feature library contains multiple baseline data feature segments of each monitoring parameter under no abnormal conditions.

[0035] In this step, the first monitoring data stream is continuously divided into multiple initial data segments according to a preset fixed time window.

[0036] Perform a pre-defined multi-dimensional feature extraction on each initial data segment to obtain the time-domain statistical features, frequency-domain transform features, and time-frequency joint features of each initial data segment;

[0037] Based on the operating condition log synchronously recorded by the power distribution terminal during the initial period, the multiple initial data segments are divided into different operating condition categories. Each operating condition category corresponds to a specific combination of load level, environmental conditions and operating mode. The operating condition log includes load rate information, ambient temperature information and equipment switching status information.

[0038] The multi-dimensional features of all initial data segments belonging to the same working condition category are combined into a high-dimensional feature vector, where one initial data segment corresponds to one high-dimensional feature vector.

[0039] Based on the preset clustering algorithm, all high-dimensional feature vectors under the same working condition category are clustered to form multiple feature clusters, and the standard deviation of the feature vectors within each feature cluster is calculated.

[0040] For target feature clusters whose standard deviation of feature vectors within a cluster is less than a preset standard deviation threshold, the feature center vector of the target feature cluster is used as a reference data feature segment and associated with the label of the working condition category to which the target feature cluster belongs.

[0041] The baseline data feature fragments under all operating condition categories are summarized to construct the normal data feature library.

[0042] Specifically, the acquired first monitoring data stream (initial learning period data) undergoes standardized preprocessing, including data alignment, invalid value removal, and noise filtering. Subsequently, the data stream is continuously and non-overlappingly segmented using fixed-duration windows (e.g., each window lasts 10 seconds and contains 50 power frequency cycles), resulting in a series of equal-length initial data segments. Each segment contains information on the synchronous changes of various monitoring parameters within a short period.

[0043] Next, a standardized multi-dimensional feature extraction pipeline is executed for each initial data segment:

[0044] Time-domain statistical feature extraction: Calculate the mean (reflecting the average level), variance or standard deviation (reflecting the degree of fluctuation), peak value (reflecting the extreme value), skewness and kurtosis (reflecting the shape of the distribution) of each parameter sequence within the segment.

[0045] Frequency domain transform feature extraction: Perform Fast Fourier Transform (FFT) on the segment to extract the fundamental amplitude, the content of major harmonics (such as the 3rd, 5th, and 7th harmonics), the total harmonic distortion (THD), as well as the spectral centroid and spectral entropy, etc., from the spectrum to characterize the periodic components and spectral structure of the signal.

[0046] Joint time-frequency domain feature extraction: Wavelet packet transform is used to analyze non-stationary signals and extract wavelet packet energy and energy entropy of specific frequency bands (such as low-frequency bands reflecting mechanical state and high-frequency bands reflecting switching actions) to capture the local time-frequency characteristics of the signal.

[0047] Simultaneously, the system accesses the operational status log synchronized with the first monitoring data stream. This log is provided by the SCADA system or environmental sensors, and key fields include:

[0048] Load factor information: For example, the percentage of active power to rated capacity of the current line.

[0049] Environmental conditions information: such as the ambient temperature of the terminal enclosure.

[0050] Equipment switching status information: such as the switching status of associated capacitor banks, the position of sectionalizing switches, and other discrete status quantities.

[0051] Based on this log information, the system defines a rule for classifying operating conditions. For example, the rule could be: "Load rate between 70% and 90%, ambient temperature between 20℃ and 30℃, and capacitor bank in operation" are defined as an independent operating condition category. According to this rule, the system automatically assigns each initial data segment to a specific operating condition category. This essentially adds an "environmental label" describing the context in which each data segment was generated.

[0052] Subsequently, feature fusion and clustering modeling are performed. For all initial data segments classified into the same operating condition category, the multidimensional features (time domain, frequency domain, and time-frequency domain) extracted from each segment are concatenated to form a unified high-dimensional feature vector. Thus, each operating condition category corresponds to a set of high-dimensional feature vectors. The system uses the DBSCAN (density-based clustering) algorithm to perform cluster analysis on this set of feature vectors within the same operating condition category. DBSCAN can automatically discover densely concentrated regions in the feature space; each dense region forms a feature cluster, representing a recurring and stable operating state pattern under that specific operating condition. Next, the standard deviation of all feature vectors within each feature cluster is calculated to quantify the degree of consistency within the cluster.

[0053] Finally, the benchmark library is generated. The system sets a preset standard deviation threshold as a standard to measure the "compactness" and representativeness of feature clusters. For target feature clusters with a standard deviation less than this threshold (i.e., clusters with high internal consistency and clear patterns), the mean of all feature vectors within the cluster is calculated to obtain the feature center vector of the cluster. This feature center vector is abstracted as a "digital fingerprint" of a typical stable operating state under this working condition, stored in the database as a benchmark data feature fragment, and associated with its corresponding working condition category label. By traversing all working condition categories and summarizing all benchmark data feature fragments that meet the conditions, a normal data feature library for subsequent comparative analysis is formed.

[0054] In one specific embodiment, a benchmark library is automatically constructed from historical normal operation data to characterize the "health fingerprint" of the distribution terminal under various operating conditions. Taking a distribution terminal that collects three-phase current (Ia, Ib, Ic), enclosure temperature (T), and bus voltage (U) as an example, the specific implementation process is as follows:

[0055] Step 1021: Data Partitioning

[0056] Assume the initial learning period is one month of stable operation after the equipment is put into service. The system segments the continuous monitoring data stream without overlap, using a fixed time window of 10 seconds. For example, 8640 data segments are generated per day (24 hours * 3600 seconds / 10 seconds), resulting in approximately 260,000 initial data segments in one month. Each segment contains time-series data of five parameters: Ia, Ib, Ic, T, and U, sampled at 100 points per second within 10 seconds (i.e., 1000 points for each parameter).

[0057] Step 1022: Multi-dimensional feature extraction (taking a single segment as an example)

[0058] For each 10-second data segment, perform the following feature extraction in parallel:

[0059] Time-domain statistical characteristics:

[0060] Calculate the average three-phase current: (mean(Ia), mean(Ib), mean(Ic)), which reflects the average load level;

[0061] Calculate the variance of the three-phase current: (var(Ia), var(Ib), var(Ic)), which reflects the load fluctuation.

[0062] Calculate the negative sequence unbalance of the three-phase current: |I2| / |I1|*100%, where I2 is the negative sequence current and I1 is the positive sequence current, reflecting the three-phase balance state;

[0063] Calculate the rate of temperature change: (T_end-T_start) / 10s, which reflects the trend of temperature rise.

[0064] Frequency domain transform characteristics:

[0065] Perform a Fast Fourier Transform (FFT) on each phase current to extract the 50Hz fundamental amplitude and the 150Hz (3rd harmonic) content.

[0066] Calculate the total harmonic distortion (THD): sqrt(sum of squares of the amplitudes of each harmonic) / fundamental amplitude, which reflects the quality of the current waveform.

[0067] Joint features in the time and frequency domains:

[0068] The current signal is decomposed into 4-level wavelet packets (using db4 wavelet) to extract the wavelet packet energy of the 0th node of the 3rd level (low frequency band, 0-6.25Hz). The energy in this frequency band may be related to the slow change of the background load.

[0069] Ultimately, each 10-second data segment is transformed into a multidimensional feature vector containing approximately 15-20 specific values.

[0070] Step 1023: Working Condition Classification and Tag Binding

[0071] Simultaneously, the system reads the operational log synchronized with the data for that month. The log contains one record per minute, for example:

[0072] Timestamp: 2023-10-01 14:05:00, Load factor: 75%, Ambient temperature: 28℃, Capacitor bank C1 status: Engaged

[0073] Each data segment is labeled with a working condition according to preset rules. For example, the rules are defined as follows:

[0074] Operating Condition A: Load rate (60%-80%) and temperature (25℃-30℃) with C1 in operation;

[0075] Operating Condition B: Load rate (80%-100%) and temperature (20℃-25℃), and C1 is deactivated;

[0076] Operating condition C: Load rate (40%-60%) and temperature > 30℃, and C1 is in operation;

[0077] Therefore, in the segment from 14:05:00 to 14:05:10 on October 1, 2023, the load rate was 75%, the temperature was 28℃, and C1 was activated, which meets the definition of operating condition A, and is therefore marked as "operating condition A".

[0078] Step 1024: Feature Clustering and Benchmark Extraction

[0079] Collect the approximately 20-dimensional feature vectors corresponding to all data segments labeled "Operating Condition A" (assuming there are 8000 segments).

[0080] The DBSCAN algorithm was used to cluster this set. The algorithm found that these vectors mainly clustered into two dense "clusters" in the feature space:

[0081] Cluster A1 contains approximately 7500 vectors, characterized by low negative order imbalance (<2%), low third harmonic content (<1%), and stable low-frequency wavelet energy. Calculating the standard deviation of all vectors in this cluster reveals that the standard deviations for each dimension are very small (e.g., all below 5% of the corresponding dimension's mean), indicating a highly stable and consistent state pattern.

[0082] Cluster A2: Contains approximately 500 vectors, characterized by a slightly higher negative order imbalance (2%-5%) and large fluctuations in the third harmonic content. Its internal standard deviation is relatively large, which may represent some transient transitional states or subtle changes in operating conditions that are not logged.

[0083] The preset standard deviation threshold is set to 7% of the mean of each dimension. Cluster A1 with a standard deviation below this threshold is identified as a valid target feature cluster; cluster A2 is filtered out because it is too dispersed.

[0084] The mean of all feature vectors in cluster A1 is calculated to obtain an "average feature vector," which is the benchmark data feature segment extracted from massive amounts of data, representing the most typical and stable operating state under "operating condition A." This feature vector is then associated with the label "operating condition A" and stored in the database.

[0085] Step 1025: Construct the feature library

[0086] Repeat step 1024 for all identified operating condition categories, such as "Operating Condition B" and "Operating Condition C". Ultimately, the system obtains a normal data feature library, which may contain dozens of entries, for example:

[0087] Item 1: [Feature Vector Va], associated tags: Operating Condition A (medium-high load, medium temperature, capacitor in operation);

[0088] Item 2: [Feature Vector Vb], associated tags: Operating Condition B (High Load, Medium to Low Temperature, Capacitor Disconnection);

[0089] Item 3: [Feature Vector Vc], associated tag: Operating condition C (medium load, high temperature, capacitor in operation).

[0090] In summary, the method for constructing a normal data feature library provided in this step enables the transformation from massive, high-dimensional, and dynamic monitoring data to a structured, interpretable, and computable state knowledge model, resulting in significant technological improvements.

[0091] This method represents a leap in condition assessment from "single-point threshold" to "multi-condition pattern matching": traditional methods rely on fixed thresholds, failing to distinguish normal fluctuations under different loads and temperatures. This method introduces condition category classification and associates it with condition labels, assigning a clear operational context to each baseline feature segment. This transforms subsequent condition judgment from simple numerical comparison to feature pattern matching within the same or similar operating condition context, significantly improving the accuracy and context adaptability of condition assessment and fundamentally reducing false alarms caused by normal changes in operating conditions.

[0092] A high-fidelity and robust "digital profile" set of normal operating states was constructed: through multi-dimensional feature extraction (time domain, frequency domain, and time-frequency domain), this method comprehensively depicts the operating state from multiple physical perspectives, avoiding the one-sidedness of information from a single feature. Combined with density-based clustering (DBSCAN) and standard deviation threshold filtering, it can automatically identify and eliminate noise and transitional data, retaining only those stable state patterns that repeatedly appear and are highly consistent under specific operating conditions as benchmarks. The resulting feature library is not a simple accumulation of all historical data, but a refined, high-quality set of "state prototypes," providing a reliable benchmark for accurate comparison and significantly improving the signal-to-noise ratio and reliability of anomaly detection.

[0093] This lays a structured knowledge foundation for the system's self-learning and dynamic evolution: the normal data feature library produced in this step has a data structure (feature vectors + operating condition labels) with inherent scalability. This is not merely a static database for comparison, but a structured knowledge framework. When the system's operating state undergoes a legitimate and persistent change (such as equipment updates or network reconstruction leading to a new stable operating point), the new state pattern can follow the same "feature extraction - operating condition classification - clustering verification" logic and be incorporated into this library as a new benchmark segment. This provides a core data foundation and feasible path for the entire monitoring system to achieve online learning, knowledge accumulation, and dynamic evolution, enabling the system to continuously adapt to the development and changes in the power distribution network and maintain long-term effectiveness.

[0094] Step S103: Obtain the second monitoring data stream of the power distribution terminal in the current time period, parse the second monitoring data stream, and obtain the abnormal data feature set and the current normal data feature set.

[0095] In this step, the second monitoring data stream is divided into multiple current data segments by using the same preset fixed time window as the first monitoring data stream.

[0096] Perform a preset multi-dimensional feature extraction on a certain current data segment, and combine a certain time-domain statistical feature, a certain frequency-domain transform feature, and a certain time-frequency domain joint feature to form a certain current multi-dimensional feature vector corresponding to a certain current data segment;

[0097] Obtain a real-time operating condition information corresponding to the acquisition time of a certain current data segment, wherein the real-time operating condition information includes at least a real-time load rate and a real-time ambient temperature;

[0098] The real-time operating condition information is compared with the operating condition categories associated with each benchmark data feature segment stored in the normal data feature library, and the operating condition distance between the real-time operating condition information and the center value of the operating condition range defined by each operating condition category is calculated.

[0099] Select the top K operating condition categories with the smallest operating condition distance, and retrieve all benchmark data feature segments associated with the top K operating condition categories from the normal data feature library as a matching candidate segment set;

[0100] For each candidate baseline data feature segment in the matching candidate segment set, the weighted Euclidean distance between a current multidimensional feature vector of a current data segment and the feature center vectors of each candidate baseline data feature segment is calculated as the dynamic matching distance. The weight coefficient is preset according to the discriminative power of each feature dimension in historical data.

[0101] Set a dynamic anomaly detection threshold. If the minimum value among all dynamic matching distances corresponding to a certain current data segment is greater than the anomaly detection threshold, then the current data segment is determined to be abnormal data, and a certain multi-dimensional feature vector is assigned to the abnormal data feature set.

[0102] If there exists at least one dynamic matching distance corresponding to a certain current data segment that is less than or equal to the dynamic anomaly determination threshold, then the certain current data segment is determined to be current normal data, and a certain multidimensional feature vector is assigned to the current normal data feature set.

[0103] In one specific embodiment, online parsing and status determination of the real-time monitoring data stream are performed. The core of this process is to utilize a pre-built normal data feature library to perform "contextualized" anomaly screening on the current data segment. The power distribution terminal monitors three-phase current (Ia, Ib, Ic), temperature (T), and voltage (U). The current time period is set to any day after the system was put into online monitoring.

[0104] Step 1031: Real-time data segmentation and feature vectorization

[0105] Access the second monitoring data stream (real-time stream). Divide the real-time stream continuously into segments using the same 10-second fixed time window as when building the feature library, resulting in individual current data fragments. For example, process a segment with timestamps from 14:30:00 to 14:30:10 on November 1, 2023.

[0106] Perform the same multi-dimensional feature extraction process as in step S102 on this segment.

[0107] Finally, the 10 seconds of raw data were transformed into a current multidimensional feature vector, for example: [mean(Ia)=85A,var(Ib)=4.2, negative order imbalance=1.8%, THD=1.5%, temperature change rate=0.1℃ / s].

[0108] Step 1032: Obtain real-time operating conditions and match candidate benchmarks

[0109] Obtain the real-time operating information at the time of acquisition (2023-11-01 14:30:05): The real-time load rate is read as 78% from the SCADA (Supervisory Control And Data Acquisition) system, the real-time ambient temperature is read as 29℃ from the temperature sensor, and it is confirmed that capacitor bank C1 is in the working state. This information is recorded as [Load rate: 78%, Temperature: 29℃, C1: In the working state].

[0110] Working condition distance calculation and candidate set generation:

[0111] Iterate through the operating condition category labels associated with each baseline data feature segment in the normal data feature library. Each label defines a multi-dimensional operating condition range, such as "Operating Condition A: Load Rate (60%-80%), Temperature (25℃-30℃), C1 Input". Calculate the center value of this range, for example, the center of the load rate is 70%, and the center of the temperature is 27.5℃.

[0112] Calculate the Euclidean distance (or Manhattan distance) between the real-time operating condition information [78%, 29℃] and the center value of each operating condition category, as the operating condition distance. For example, the distance to the center of "Operating Condition A" might be sqrt((78-70)^2+(29-27.5)^2)≈8.14.

[0113] With K=3 as the default, select the three working condition categories with the smallest working condition distance (such as "Working Condition A", "Working Condition D", "Working Condition F").

[0114] From the feature library, all baseline data feature fragments associated with these three categories are retrieved to form a matching candidate fragment set. This set contains those "normal state fingerprints" that are most similar to the current operating conditions in historical data.

[0115] Step 1033: Weighted distance calculation and dynamic threshold determination

[0116] Weighted Euclidean distance calculation: For each candidate baseline segment in the matching candidate segment set (e.g., the baseline feature vector Va representing "condition A"), calculate the weighted Euclidean distance between the feature vector of the current data segment and Va.

[0117] Preset weighting coefficients: During system initialization, historical normal data is analyzed to calculate the discriminative power (F-value) of each feature dimension (e.g., "negative order imbalance" or "THD") in distinguishing different steady states (e.g., using ANOVA). Features with high discriminative power (e.g., "THD" is sensitive to arc faults) are assigned higher weights (e.g., 0.2); features with low discriminative power (e.g., "mean current" varies normally with load) are assigned lower weights (e.g., 0.05). All weights are normalized and summed to 1.

[0118] Dynamic anomaly detection threshold setting: This threshold is not a fixed value.

[0119] Benchmark value: Based on historical statistics, a basic threshold T is determined. h .

[0120] Dynamic adjustment: Dynamic anomaly detection threshold = T h *α, where the adjustment factor α is a function of the real-time load rate and the ambient temperature. For example, the rule can be set as follows: when the real-time load rate > 85% or the ambient temperature > 35°C, α = 1.2 (relaxing the threshold because normal fluctuations increase under extreme conditions); otherwise, α = 1.0. This allows the judgment criterion to adapt to the current stress level of the system.

[0121] Step 1034: State determination and set merging

[0122] Regarding the data segment currently being processed:

[0123] Anomaly detection: If the minimum value among all calculated dynamic matching distances (i.e., the distance to the most similar historical normal pattern) is still greater than the dynamic anomaly detection threshold, then it is determined that even under the most similar working conditions, the current state deviates too far from the normal pattern. Therefore, this segment is identified as anomalous data, and its corresponding multidimensional feature vector is included in the anomalous data feature set.

[0124] Normal state determination: If at least one dynamic matching distance is less than or equal to the dynamic anomaly determination threshold, it indicates that the current state sufficiently matches at least one historical normal pattern. Therefore, the segment is determined to be current normal data, and its multidimensional feature vector is included in the current normal data feature set.

[0125] For each current data segment in the real-time stream, repeat steps 1031 to 1034 above to continuously parse and classify the real-time data stream into two feature sets, providing input for subsequent correlation analysis and decision-making.

[0126] In summary, the real-time analysis and status determination method provided in this step, by introducing a context-aware matching mechanism and adaptive dynamic thresholds, achieves accurate and efficient online monitoring of the operating status of power distribution terminals, bringing significant technical improvements.

[0127] This method achieves "precise guidance" in anomaly detection, significantly reducing false alarms: Traditional methods use a globally fixed threshold, which cannot distinguish between "normal high current under high load" and "abnormal high current under low load." This method, through real-time operating condition matching, intelligently selects the most relevant historical normal pattern as the comparison benchmark for each data segment, achieving "comparison within the same category." Combined with weighted Euclidean distance (highlighting key anomaly features), it ensures that the judgment is based on the degree of pattern deviation under the same operating conditions, rather than absolute numerical magnitude. This allows the system to accurately identify true pattern anomalies, rather than normal numerical fluctuations caused by operating conditions.

[0128] A flexible criterion adapted to dynamic system pressure was constructed to improve robustness: by introducing a dynamic anomaly detection threshold, this method makes the anomaly detection standard no longer a rigid "red line". When the system is under high load, high temperature and other pressure conditions, the detection threshold is automatically and appropriately relaxed to tolerate the increase in normal fluctuations caused by the physical characteristics of the equipment; under stable operating conditions, a strict standard is maintained. This flexible mechanism effectively avoids generating a large number of interfering alarms ("alarm storm") due to excessive sensitivity when the system is most vulnerable and requires the most attention, ensuring the stability and practicality of the monitoring system under different operating pressures.

[0129] This creates a "high-quality data pool" for subsequent intelligent analysis, improving analytical efficiency. The outputs of this step—the set of abnormal data features and the set of current normal data features—are not simply data aggregations. The abnormal set is a pre-filtered, high-confidence feature library of suspicious events; the current normal set is data that has been filtered to remove obvious anomalies and represents the actual steady-state operating characteristics under various current conditions. These two structured and labeled sets provide high-quality, highly targeted input for subsequent steps (such as the correlation analysis and knowledge base update in step S105), avoiding the huge overhead of full-data processing. This makes the entire analysis process focused and computationally efficient, enabling real-time or near-real-time intelligent decision-making.

[0130] Step S104: Determine whether there exists a reference data feature segment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set.

[0131] In one specific embodiment, after determining whether there exists a reference data feature segment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set, if there exists a reference data feature segment that matches a certain abnormal data feature in the abnormal data feature set, then the feature offset between the certain abnormal data feature and the reference data feature segment is calculated; based on the magnitude of the feature offset, the severity level of the abnormal event corresponding to the certain abnormal data feature is determined; if the severity level exceeds a preset alarm threshold, then alarm information containing a certain abnormal data feature, a certain reference data feature segment, and the feature offset is immediately generated and pushed to the user terminal.

[0132] After initially separating out the abnormal data, a crucial "rapid diagnosis" is performed to determine whether the current anomaly belongs to a type already recognized by the system, and to quantitatively assess known anomalies. The implementation process is as follows:

[0133] Iterate through each anomalous data feature (i.e., a multidimensional feature vector representing an anomalous segment) in the anomalous data feature set. For a specific anomalous data feature A being processed... exp The system performs the following operations:

[0134] Quick match query: will not match A exp Instead of performing a full comparison with all benchmarks in the normal data feature library (as done in step S103), it performs a more precise targeted matching. It uses an efficient similarity retrieval algorithm (such as one based on Locality Sensitive Hashing (LSH) or a constructed feature vector index) to quickly search the normal data feature library for the existence of a benchmark data feature fragment B. ref This causes the matching degree between the two to exceed a high "known abnormal matching threshold".

[0135] Matching degree calculation: Cosine similarity or an improved similarity that considers feature weights is typically used. The formula is: Matching degree = sim(A exp B ref This threshold (e.g., 0.85) is set more stringently than the dynamic anomaly determination threshold in step S103, and is intended to filter out highly similar known patterns.

[0136] Calculate the feature offset: If such a B exists ref (i.e., a successful match) the system determines that this anomaly belongs to the "known anomaly pattern". Next, A is calculated. exp With B ref The characteristic offset Δ between them.

[0137] Severity level determination: A severity level mapping table is predefined. This table classifies anomalous events into multiple levels based on the range of feature offsets Δ, for example:

[0138] Level 1 (hint): Δ is between thresholds T1 and T2 (e.g., slightly deviating from the baseline).

[0139] Level 2 (Warning): Δ is between the thresholds T2 and T3 (e.g., moderate deviation).

[0140] Level 3 (Severe): Δ is greater than the threshold T3 (e.g., severely deviated).

[0141] Thresholds T1, T2, and T3 can be set by analyzing historical known anomaly cases, and can be tailored to different B values. ref The exception type it represents can be customized.

[0142] Alarm Generation and Push: The system continuously checks whether the calculated severity level exceeds a preset alarm threshold (e.g., level 2 and above require immediate alarm). If it does, a structured alarm message is automatically generated and pushed to the user's device. This alarm message includes:

[0143] Core data: Anomalous data characteristics that trigger alarms exp A summary or related index of the original data fragments.

[0144] Matching benchmark: Identified benchmark data feature fragment B ref The identifier and its associated known anomaly type description (such as "overload characteristic mode" or "three-phase imbalance characteristic mode").

[0145] Quantitative assessment: The calculated feature offset Δ value and severity level.

[0146] Contextual information: time of anomaly occurrence, relevant equipment identification, and real-time operating status information.

[0147] Step S105: If it does not exist, then search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature segment in the normal data feature library.

[0148] In this step, the start timestamp of the current data segment corresponding to a certain abnormal data feature is obtained; in the set of current normal data features, all current normal data features whose timestamps are within a first preset time window before the start timestamp are searched as candidate precursor association features; in the set of current normal data features, all current normal data features whose timestamps are within a second preset time window after the start timestamp are searched as candidate successor association features, wherein the lengths of the first preset time window and the second preset time window are dynamically set according to the estimated duration of the abnormal event corresponding to the abnormal data feature; the correlation coefficient between a certain abnormal data feature and each candidate precursor association feature and candidate successor association feature in the feature space is calculated; one or more current normal data features with a correlation coefficient greater than a preset association threshold are selected as current normal data features associated with the certain abnormal data feature.

[0149] Further, first real-time operating condition information corresponding to the first current normal data feature is obtained, wherein the first current normal data feature is any one of the at least one current normal data features; from the normal data feature library, all first benchmark data feature segments of the same operating condition category as the first real-time operating condition information are retrieved; the weighted similarity between the first feature vector of the first current normal data feature and the feature center vector of each first benchmark data feature segment is calculated; and each weighted similarity is fused according to a preset weighted average function to obtain a first comprehensive matching degree value, which is used as the matching degree of the first current normal data feature.

[0150] In one specific embodiment, the core intelligent analysis process is executed when an unknown type of anomaly is identified (no known benchmark was matched in step S104). Its goal is to determine whether this anomaly indicates a new, legitimate system state transition by analyzing the normal operating states adjacent to the anomaly event, thereby deciding whether to expand the system's cognitive database. The specific implementation is as follows:

[0151] Part 1: Finding Normal Characteristics of Associations

[0152] Determine the analysis object and time base: Assume the system is processing a feature vector A from an anomaly data feature set. unknown This corresponds to an unknown anomalous segment occurring at time TA. The system first obtains the start timestamp T of this segment. start .

[0153] Dynamically set the associated time window: based on A unknown The characteristics (such as abnormal amplitude and rate of change) are used to estimate the possible duration of the abnormal event. eventFor example, a rule could be set: if the total harmonic distortion of the current suddenly increases by more than 10%, then the predicted D... event = 30 seconds; if it is only a small temperature fluctuation, then D event =10 seconds. Then, set the first preset time window W. pre =2*D event (e.g., 60 seconds before an anomaly), the second preset time window W post =D event (e.g., 30 seconds after an abnormality).

[0154] Retrieve candidate related features: Perform two searches within the current set of normal data features:

[0155] The retrieval timestamp falls on [T] start -W pre T start All current normal data features within the interval are used as the candidate precursor association feature set {P}. i}

[0156] The retrieval timestamp falls on [T] start +D event T start +D event +W post All current normal data features within the interval are used as candidate successor association feature set {S}. j}

[0157] Calculate relevance and filter: Calculate A unknown With each candidate feature P i S j The correlation coefficient in the feature space. Here, we use the cosine similarity of the feature vectors and consider the time decay factor:

[0158] Correlation coefficient P i =cos sim (A unknown P i )*exp(-λ*(T start -TP i ), where cos sim (A unknown P i ) represents the cosine similarity between the feature vector of the anomalous data and the normal feature vector of the candidate predecessor, TP i It is P i The timestamp, λ is the decay coefficient, which makes features that are closer to the anomaly in time have higher weights.

[0159] Similarly, calculate the correlation coefficient S. j .

[0160] Set a preset correlation threshold (e.g., 0.6). Filter out all P-values ​​with a correlation coefficient greater than this threshold. i and S j They are considered to be related to A unknown The current normal data features that are closely related in feature evolution are denoted as the associated feature set {R}. k}

[0161] Part Two: Calculating the matching degree of associated normal features

[0162] For the associated feature set {R} k Each feature in} (denoted as the first current normal data feature R1):

[0163] Determine the comparison benchmark: Obtain the first real-time operating condition information corresponding to R1 (e.g., load rate 75%, temperature 28℃). Then, retrieve all first benchmark data feature fragments from the normal data feature library whose operating condition category labels are completely identical to this information, forming a benchmark set {B}. m}

[0164] Calculate weighted similarity: Calculate the eigenvectors of R1 and the reference set {B} m The weighted cosine similarity between the feature center vectors of each reference in}.

[0165] Weighted similarity (R1, B) m )=∑(w d *(R 1，d *B m，d )) / (R1|| w *||B m || w ), where R 1，d Let B be the eigenvalue of vector R1 in the d-th dimension. m，d For vector B m The eigenvalues ​​in the d-th dimension, B m Let be the feature center vector of the m-th baseline data feature segment, d represent the feature dimension, and ∑ represent the summation over all feature dimensions d, ||R1|| w Let ||B| be the weighted norm of vector R1. m || w For vector B m The weighted norm, W d is the weight coefficient of the d-th feature dimension. The weight wd is calculated based on the ability of this dimension to distinguish different steady states in historical data (such as the F-value of ANOVA). The higher the discrimination, the greater the weight.

[0166] The fusion yields a comprehensive matching degree: a set of weighted similarity values ​​{WS}. m}. Calculate the first comprehensive matching score M1 as the matching score of R1. M1 is a measure of the matching score of {WS}. m The weighted average of}

[0167] M1=Σ(α m *WS m ) / Σα m ,

[0168] WS m Let α be the m-th weighted similarity value. m It is vector B m The relevance of the operating conditions to the current real-time operating conditions can be expressed by comparing the operating condition information with B. m It is represented by the reciprocal of the distance to the center value of the corresponding operating condition category. This gives historical benchmarks that are more closely aligned with the current operating conditions greater weight in the overall score.

[0169] Calculate the matching degree for all associated features: for the associated feature set {R} k Repeat the above process for each associated feature in} to obtain a set of matching degree values ​​{M}. k Typically, the average or minimum value of these matching degrees is taken as the final matching degree representing the entire associated feature set, and used for the threshold determination in step S106.

[0170] Step S106: Determine whether the matching degree is greater than a preset threshold.

[0171] In one specific embodiment, after determining whether the matching degree is greater than a preset threshold, if it is not greater than the preset threshold, then the data segment corresponding to the at least one current normal data feature is not added to the normal data feature library.

[0172] Step S107: If the value is greater than a preset threshold, the data segment corresponding to the at least one current normal data feature is added to the normal data feature library, and the updated normal data feature library and the abnormal data features associated with the at least one current normal data feature are transmitted to the user terminal.

[0173] In one specific embodiment, after determining in step S106 that the overall matching degree is greater than a preset threshold, it is confirmed that the normal operating state associated with the abnormal event represents a new and legitimate system steady state, which needs to be incorporated into the knowledge system, and complete diagnostic information should be provided to the user. The specific implementation method is as follows:

[0174] Part 1: Knowledge Base Update Processing

[0175] Generation of new baseline data segments:

[0176] For abnormal data features A determined by threshold values unknown Related feature set {R kThe system performs the following steps to generate a new baseline:

[0177] Feature fusion: For the associated feature set {R} k All eigenvectors R in} k For each feature dimension d, calculate its weighted average R. new，d The weights are based on each R. k Anomalous data feature A unknown The temporal proximity and correlation coefficient settings give greater weight to features that are closer to the anomaly in time and have a higher correlation.

[0178] Consistency verification: Calculate the new fused feature vector R new With the associated feature set {R k The weighted Euclidean distance of each vector in the expression is used to verify whether it falls within the preset consistency tolerance range. If it exceeds the range, the generation process is abandoned, and a manual review process is triggered.

[0179] Generate baseline vectors: After verification, R... new The feature center vector serves as the new benchmark.

[0180] Determination of operating condition labels and characteristic distribution parameters:

[0181] Operating condition tags: Acquisition and {R k The corresponding real-time operating condition information (such as load rate, temperature, etc.) is used as the operating condition category label for the new benchmark.

[0182] Feature distribution range: based on {R k The standard deviation of all feature vectors in each dimension is multiplied by a preset expansion coefficient (e.g., 1.5) to generate the feature distribution range parameters of the new benchmark. This defines the boundary for determining whether future data matches this benchmark.

[0183] Knowledge base structured storage

[0184] The newly generated benchmark entries are stored in a structured format in the normal data feature library:

[0185] Base ID: An automatically generated unique identifier (such as UUID);

[0186] Feature center vector: R new ;

[0187] Feature distribution range: standard deviation of each dimension;

[0188] Operating condition tags: Associated operating condition descriptions;

[0189] Generate timestamp: Current system time;

[0190] Associated Exception ID: The identifier of the exception event that triggered this update;

[0191] Confidence metrics: based on matching degree and {R k Consistency calculation of}.

[0192] This step, by transmitting the updated normal data feature library and the abnormal data features associated with at least one current normal data feature to the user terminal, goes beyond a simple alarm; it delivers a structured report containing a complete chain of evidence. Operations personnel not only know "what happened," but also understand "why the system made that judgment" and "what new knowledge they have learned." This transparent design enhances user trust in the system's automated decision-making and provides ample evidence for manual review.

[0193] In summary, the method presented in this application achieves accurate and adaptive monitoring of the operating status of power distribution terminals by constructing a normal data feature library dynamically associated with multi-dimensional operating conditions and employing weighted similarity matching and dynamic anomaly judgment thresholds. Furthermore, by deeply analyzing the matching degree between normal data features associated with unknown anomaly events and historical benchmarks, it intelligently decides whether to include feature fragments representing new steady states in the feature library, forming a complete technical closed loop of data parsing, anomaly identification, association analysis, and knowledge self-evolution. This method effectively solves the industry problems of high false alarm rates and inability to adapt to complex and changing operating conditions caused by traditional fixed threshold alarms, significantly improving the accuracy of state assessment and context awareness. The dynamic update mechanism of the knowledge base enables the system to continuously learn and evolve, automatically tracking and learning legitimate new states of the power distribution system generated by equipment aging, network reconstruction, or operational mode optimization, achieving a leap from a "static rule base" to a "dynamic intelligent agent," thereby significantly reducing operation and maintenance costs and improving the timeliness and reliability of fault warnings. This provides core technical support for building a highly reliable intelligent power distribution network with predictive maintenance capabilities.

[0194] Please see Figure 2 The diagram shows a structural block diagram of a power distribution terminal monitoring information analysis system according to this application.

[0195] like Figure 2 As shown, the monitoring information analysis system 200 includes an acquisition module 210, a first parsing module 220, a second parsing module 230, a first judgment module 240, a calculation module 250, a second judgment module 260, and an update module 270.

[0196] The acquisition module 210 is configured to acquire the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period; the first parsing module 220 is configured to parse the first monitoring data stream and construct a normal data feature library, wherein the normal data feature library contains multiple base data feature segments of each monitoring parameter under no-abnormal conditions; the second parsing module 230 is configured to acquire the second monitoring data stream of the power distribution terminal during the current time period, parse the second monitoring data stream, and obtain an abnormal data feature set and a current normal data feature set; the first judgment module 240 is configured to judge whether there is a base data feature in the normal data feature library that matches any abnormal data feature in the abnormal data feature set. The system includes: a quasi-data feature fragment; a calculation module 250 configured to, if not present, search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library; a second judgment module 260 configured to determine whether the matching degree is greater than a preset threshold; and an update module 270 configured to, if greater than the preset threshold, add the data fragment corresponding to the at least one current normal data feature to the normal data feature library, and transmit the updated normal data feature library and the abnormal data feature associated with the at least one current normal data feature to the user terminal.

[0197] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0198] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the monitoring information analysis method for the power distribution terminal in any of the above method embodiments.

[0199] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0200] Acquire the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period;

[0201] The first monitoring data stream is parsed to construct a normal data feature library, which contains multiple baseline data feature fragments of each monitoring parameter under no abnormal conditions.

[0202] Acquire the second monitoring data stream of the power distribution terminal in the current time period, parse the second monitoring data stream to obtain the abnormal data feature set and the current normal data feature set;

[0203] Determine whether there exists a benchmark data feature fragment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set;

[0204] If it does not exist, then search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library.

[0205] Determine whether the matching degree is greater than a preset threshold;

[0206] If the value exceeds a preset threshold, the data segment corresponding to the at least one current normal data feature is added to the normal data feature library, and the updated normal data feature library and the abnormal data features associated with the at least one current normal data feature are transmitted to the user terminal.

[0207] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the power distribution terminal's monitoring information analysis system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely configured relative to a processor, and these remote memories can be connected to the power distribution terminal's monitoring information analysis system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0208] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the power distribution terminal monitoring information analysis method described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the power distribution terminal monitoring information analysis system. The output device 340 may include a display screen or other display device.

[0209] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0210] In one implementation, the above-described electronic device is applied to a monitoring information analysis system for a power distribution terminal, serving as a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0211] Acquire the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period;

[0212] The first monitoring data stream is parsed to construct a normal data feature library, which contains multiple baseline data feature fragments of each monitoring parameter under no abnormal conditions.

[0213] Acquire the second monitoring data stream of the power distribution terminal in the current time period, parse the second monitoring data stream to obtain the abnormal data feature set and the current normal data feature set;

[0214] Determine whether there exists a benchmark data feature fragment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set;

[0215] If it does not exist, then search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library.

[0216] Determine whether the matching degree is greater than a preset threshold;

[0217] If the value exceeds a preset threshold, the data segment corresponding to the at least one current normal data feature is added to the normal data feature library, and the updated normal data feature library and the abnormal data features associated with the at least one current normal data feature are transmitted to the user terminal.

[0218] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0219] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for analyzing monitoring information of a power distribution terminal, characterized in that, include: Acquire the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period; The first monitoring data stream is parsed to construct a normal data feature library, which contains multiple baseline data feature fragments of each monitoring parameter under no abnormal conditions. Acquire the second monitoring data stream of the power distribution terminal in the current time period, parse the second monitoring data stream to obtain the abnormal data feature set and the current normal data feature set; Determine whether there exists a benchmark data feature fragment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set; If it does not exist, then search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library. Determine whether the matching degree is greater than a preset threshold; If the value exceeds a preset threshold, the data segment corresponding to the at least one current normal data feature is added to the normal data feature library, and the updated normal data feature library and the abnormal data features associated with the at least one current normal data feature are transmitted to the user terminal.

2. The method for analyzing monitoring information of a power distribution terminal according to claim 1, characterized in that, The step of parsing the first monitoring data stream and constructing a normal data feature library includes: The first monitoring data stream is continuously divided into multiple initial data segments according to a preset fixed time window. Perform a pre-defined multi-dimensional feature extraction on each initial data segment to obtain the time-domain statistical features, frequency-domain transform features, and time-frequency joint features of each initial data segment; Based on the operating condition log synchronously recorded by the power distribution terminal during the initial period, the multiple initial data segments are divided into different operating condition categories. Each operating condition category corresponds to a specific combination of load level, environmental conditions and operating mode. The operating condition log includes load rate information, ambient temperature information and equipment switching status information. The multi-dimensional features of all initial data segments belonging to the same working condition category are combined into a high-dimensional feature vector, where one initial data segment corresponds to one high-dimensional feature vector. Based on the preset clustering algorithm, all high-dimensional feature vectors under the same working condition category are clustered to form multiple feature clusters, and the standard deviation of the feature vectors within each feature cluster is calculated. For target feature clusters whose standard deviation of feature vectors within a cluster is less than a preset standard deviation threshold, the feature center vector of the target feature cluster is used as a reference data feature segment and associated with the label of the working condition category to which the target feature cluster belongs. The baseline data feature fragments under all operating condition categories are summarized to construct the normal data feature library.

3. The method for analyzing monitoring information of a power distribution terminal according to claim 2, characterized in that, The step of acquiring the second monitoring data stream of the power distribution terminal in the current time period, parsing the second monitoring data stream, and obtaining the abnormal data feature set and the current normal data feature set includes: The second monitoring data stream is divided into multiple current data segments by using the same preset fixed time window as the first monitoring data stream. Perform a preset multi-dimensional feature extraction on a certain current data segment, and combine a certain time-domain statistical feature, a certain frequency-domain transform feature, and a certain time-frequency domain joint feature to form a certain current multi-dimensional feature vector corresponding to a certain current data segment; Obtain a real-time operating condition information corresponding to the acquisition time of a certain current data segment, wherein the real-time operating condition information includes at least a real-time load rate and a real-time ambient temperature; The real-time operating condition information is compared with the operating condition categories associated with each benchmark data feature segment stored in the normal data feature library, and the operating condition distance between the real-time operating condition information and the center value of the operating condition range defined by each operating condition category is calculated. Select the top K operating condition categories with the smallest operating condition distance, and retrieve all benchmark data feature segments associated with the top K operating condition categories from the normal data feature library as a matching candidate segment set; For each candidate baseline data feature segment in the matching candidate segment set, the weighted Euclidean distance between a current multidimensional feature vector of a current data segment and the feature center vectors of each candidate baseline data feature segment is calculated as the dynamic matching distance. The weight coefficient is preset according to the discriminative power of each feature dimension in historical data. Set a dynamic anomaly detection threshold. If the minimum value among all dynamic matching distances corresponding to a certain current data segment is greater than the anomaly detection threshold, then the current data segment is determined to be abnormal data, and a certain multi-dimensional feature vector is assigned to the abnormal data feature set. If there exists at least one dynamic matching distance corresponding to a certain current data segment that is less than or equal to the dynamic anomaly determination threshold, then the certain current data segment is determined to be current normal data, and a certain multidimensional feature vector is assigned to the current normal data feature set.

4. The method for analyzing monitoring information of a power distribution terminal according to claim 1, characterized in that, After determining whether a baseline data feature fragment exists in the normal data feature library that matches any abnormal data feature in the abnormal data feature set, the method further includes: If there exists a reference data feature segment that matches a certain abnormal data feature in the abnormal data feature set, then calculate the feature offset between the abnormal data feature and the reference data feature segment. Based on the magnitude of the feature offset, the severity level of the abnormal event corresponding to a certain abnormal data feature is determined; If the severity level exceeds the preset alarm threshold, an alarm message containing a certain abnormal data feature, a certain baseline data feature segment, and feature offset will be immediately generated and pushed to the user terminal.

5. The method for analyzing monitoring information of a power distribution terminal according to claim 1, characterized in that, The step of searching for at least one current normal data feature associated with any of the abnormal data features in the current normal data feature set includes: Get the start timestamp of the current data segment corresponding to a certain abnormal data feature; In the set of current normal data features, find all current normal data features whose timestamps are within a first preset time window before the start timestamp, and use them as candidate precursor association features; In the set of current normal data features, all current normal data features whose timestamps are located within a second preset time window after the start timestamp are searched as candidate subsequent associated features. The lengths of the first preset time window and the second preset time window are dynamically set according to the estimated duration of the abnormal event corresponding to the abnormal data feature. Calculate the correlation coefficient between a certain abnormal data feature and each candidate predecessor association feature and candidate successor association feature in the feature space; One or more current normal data features with a correlation coefficient greater than a preset association threshold are selected as current normal data features associated with the aforementioned abnormal data feature.

6. The method for analyzing monitoring information of a power distribution terminal according to claim 1, characterized in that, The calculation of the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library includes: Obtain first real-time operating condition information corresponding to a first current normal data feature, wherein the first current normal data feature is any one of the at least one current normal data features; From the normal data feature library, retrieve all first baseline data feature fragments that are of the same working condition category as the first real-time working condition information; Calculate the weighted similarity between the first feature vector of the first current normal data feature and the feature center vectors of each first benchmark data feature segment; The weighted similarities are fused according to a preset weighted average function to obtain a first comprehensive matching degree value, which is used as the matching degree of the first current normal data feature.

7. The method for analyzing monitoring information of a power distribution terminal according to claim 1, characterized in that, After determining whether the matching degree is greater than a preset threshold, the method further includes: If the value is not greater than a preset threshold, the data segment corresponding to at least one current normal data feature will not be added to the normal data feature library.

8. A monitoring information analysis system for a power distribution terminal, characterized in that, include: The acquisition module is configured to acquire the first monitoring data stream of each monitoring parameter of the power distribution terminal during the initial time period; The first parsing module is configured to parse the first monitoring data stream and construct a normal data feature library, wherein the normal data feature library contains multiple baseline data feature fragments of each monitoring parameter under no abnormal conditions. The second parsing module is configured to acquire the second monitoring data stream of the power distribution terminal in the current time period, parse the second monitoring data stream, and obtain the abnormal data feature set and the current normal data feature set; The first judgment module is configured to determine whether there is a benchmark data feature fragment in the normal data feature library that matches any abnormal data feature in the abnormal data feature set; The calculation module is configured to, if it does not exist, search for at least one current normal data feature associated with any abnormal data feature in the current normal data feature set, and calculate the matching degree between the at least one current normal data feature and each benchmark data feature fragment in the normal data feature library. The second judgment module is configured to determine whether the matching degree is greater than a preset threshold. The update module is configured to add the data segment corresponding to the at least one current normal data feature to the normal data feature library if the value is greater than a preset threshold, and transmit the updated normal data feature library and the abnormal data features associated with the at least one current normal data feature to the user terminal.

9. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 7.

Images