An intelligent analysis system for power monitoring data based on a mutual inductor

Abstract

This invention relates to the field of power monitoring and analysis technology, and discloses an intelligent power monitoring data analysis system based on instrument transformers. The system includes an instrument transformer data acquisition and structuring module, which acquires current waveforms, voltage waveforms, and harmonic components in real time, and generates standardized monitoring data units through preprocessing, field mapping, and association identification. A theoretical monitoring value calculation module constructs a dynamic calculation model using historical power data, inputting standardized data units and outputting theoretical values. A rule compliance verification module matches equipment type and operating rules to generate a single compliance judgment. A multi-dimensional difference analysis module compares theoretical and measured values ​​from time-domain, frequency-domain deviations, and waveform distortion to generate an equipment-level difference coefficient matrix. An association network construction module builds a multi-monitoring point association map based on equipment identification and location information. An anomaly location and strategy generation module combines the map, compliance judgment, and difference matrix to calculate risk scores, generate anomaly probability distribution maps, locate anomalies, and assign monitoring strategies.

Description

Technical Field

[0001] This invention relates to the field of power monitoring and analysis technology, specifically to an intelligent analysis system for power monitoring data based on instrument transformers. Background Technology

[0002] Accurate monitoring and effective analysis of power parameters are crucial for ensuring the stable and safe operation of power systems. Currently, power monitoring largely relies on instrument transformers to collect basic data such as current and voltage, but there are many limitations in data processing and application.

[0003] Data acquisition often lacks effective structured processing. The raw data, such as current waveforms, voltage waveforms, and harmonic components, are fragmented and lack clear mapping with preset monitoring fields or standardized data units. This makes it difficult to integrate and compare data from different devices and monitoring points, hindering efficient subsequent analysis.

[0004] Theoretical monitoring value calculations lack dynamic adaptability. Existing technologies mostly use fixed models or static formulas to calculate theoretical values, failing to fully incorporate the changing patterns of historical power data to construct dynamic calculation models. This results in a low degree of matching between theoretical values ​​and actual operating conditions, making it difficult to accurately reflect the true situation of the power system under different loads and operating states, and thus unreliable as a basis for judging whether the system is operating normally.

[0005] The methods for verifying compliance with regulations are relatively simplistic. Traditional verification processes often rely on simply comparing monitoring data against general operating rules, without accurately matching specific operating rule clauses to different types of power equipment. This results in insufficient accuracy in determining the compliance status of individual items, making it prone to misjudgments or omissions, and failing to promptly identify potential problems in equipment operation.

[0006] The dimensions of the difference analysis are not comprehensive. When comparing theoretical and measured values, existing technologies often only focus on the difference in numerical magnitude, ignoring multi-dimensional characteristics such as time domain deviation, frequency domain deviation, and waveform distortion. This makes it difficult to fully reveal potential anomalies in power system operation and fails to provide sufficient analytical basis for subsequent anomaly localization.

[0007] Insufficient correlation between data from multiple monitoring points. The data from each monitoring point are independent of each other, and no effective correlation has been established through key elements such as equipment identification and location information. As a result, it is impossible to form a complete multi-monitoring point correlation map. When an anomaly occurs in the system, it is difficult to quickly trace the source of the anomaly or accurately determine the scope of its impact, leading to low efficiency in anomaly localization.

[0008] There is a lack of systematic strategies for handling anomalies. After an anomaly is detected, most systems can only provide simple anomaly alerts, without developing targeted monitoring strategies based on risk scores and anomaly probability distribution maps. They also fail to take differentiated response measures according to the severity and distribution of the anomaly, which hinders timely control of the anomaly and ensures the stable operation of the power system. Summary of the Invention

[0009] The purpose of this invention is to provide an intelligent analysis system for power monitoring data based on current transformers, so as to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides an intelligent analysis system for power monitoring data based on instrument transformers, the system comprising:

[0011] Instrument transformer data acquisition and structuring module: Based on the real-time acquisition of power monitoring data from instrument transformer equipment, including current waveform, voltage waveform and harmonic components, the module performs signal preprocessing and field mapping, maps the data to preset monitoring fields, establishes inter-field association identifiers, and generates standardized monitoring data units.

[0012] Theoretical monitoring value calculation module: Based on historical power data, a dynamic calculation model is constructed. The standardized monitoring data unit is input into the dynamic calculation model, and the theoretical monitoring value is output.

[0013] Rule compliance verification module: Based on the standardized monitoring data unit, it retrieves data from the database to match the type of power equipment with the operating rule clauses, and generates a single compliance status judgment;

[0014] Multidimensional difference analysis module: Performs multidimensional difference analysis on the theoretical monitoring value and the measured monitoring value. The multidimensional difference analysis includes time domain deviation, frequency domain deviation and waveform distortion, and generates a device-level difference coefficient matrix.

[0015] The associated network construction module: Based on multiple standardized monitoring data units, it searches for shared device identifiers and location information among different monitoring points, constructs a graph structure with monitoring points as nodes and shared relationships as edges, and establishes a multi-monitoring point association graph;

[0016] Anomaly location and strategy generation module: Based on the multi-monitoring point association map, combined with the single compliance status judgment and the difference coefficient matrix, calculate the risk score, generate an anomaly probability distribution map, locate the anomaly area, and configure the monitoring strategy according to the anomaly probability distribution map.

[0017] Preferably, the current transformer data acquisition and structuring module specifically includes:

[0018] The data acquisition unit collects raw power data through the current transformer, performs filtering and normalization processing, and generates a preprocessed data sequence.

[0019] The field mapping unit identifies feature points in the preprocessed data sequence and maps them to preset monitoring fields, including device number, current value, voltage value, and harmonic content, to generate an original field set.

[0020] The structured processing unit performs format conversion and field reconstruction on the original field set to generate a standardized mapping field set;

[0021] The associated identification unit performs internal association matching on fields in the standardized mapping field set, establishes association relationship identifiers, and generates standardized monitoring data units.

[0022] Preferably, the theoretical monitoring value calculation module specifically includes:

[0023] Historical data feature extraction involves multi-dimensional decomposition of historical power monitoring data, including using filtering algorithms to extract waveform features, establishing the correlation between harmonics and load through harmonic analysis, and aligning waveform patterns using sequence analysis algorithms.

[0024] The dynamic calculation model is constructed by inputting the processed historical data into the prediction model. The prediction model includes a time-series prediction component based on equipment aging to generate basic prediction values, a neural network with an embedded correction mechanism to correct measurement deviations, and a real-time feature adjustment component to dynamically adjust the prediction based on real-time data.

[0025] Theoretical value calculation involves inputting real-time standardized monitoring data units into the dynamic calculation model to obtain theoretical monitoring values.

[0026] Preferably, the rule compliance verification module specifically includes:

[0027] Rule matching is based on the equipment type field and operating status field in the standardized monitoring data unit. The field values ​​are extracted and matched with the rule base to fill in missing fields, unify the expression method, and obtain a standardized set of field combinations.

[0028] The rule complexity calculation involves retrieving matching rule clauses based on a set of normalized field combinations, extracting the clause's nesting level, number of logical judgments, number of references, time range, and geographical coverage, and then calculating the rule complexity factor.

[0029] The compliance determination is based on the rule complexity factor. It calls the monitoring data and clause constraint values ​​one by one, performs condition threshold comparison, numerical range verification and timeliness judgment, and generates a single compliance status determination.

[0030] Preferably, the multidimensional difference analysis module specifically includes:

[0031] The time-domain deviation calculation involves sliding comparison of theoretical and measured monitoring values ​​using a time window, aligning sequences using a sequence alignment algorithm, calculating cumulative deviation, and generating a time-domain deviation vector.

[0032] Frequency domain deviation detection involves performing frequency domain analysis on the harmonic components of theoretical and measured values, calculating the energy spectral density ratio, extracting the harmonic deviation index, and constructing a frequency domain deviation vector.

[0033] Waveform distortion assessment is based on a matching algorithm that compares the theoretical values ​​with the morphology of the measured waveform sequences, calculates the degree of waveform distortion, and generates a distortion vector.

[0034] The difference coefficient matrix is ​​generated by concatenating the time-domain deviation vector, frequency-domain deviation vector, and distortion vector, and then outputting the difference coefficient matrix through weighted normalization.

[0035] Preferably, the associated network construction module specifically includes:

[0036] Shared entity extraction is based on multiple standardized monitoring data units. The equipment identification field and location field in each unit are extracted, and entity classification, matching and deduplication are performed to generate a shared entity set.

[0037] The correlation density calculation involves statistically analyzing the frequency and category of shared entities at different monitoring points, recording distribution information, and calculating the correlation density index.

[0038] Graph structure construction: Based on the association density index, the monitoring points are coded into nodes, the shared relationship strength value between nodes is calculated, and a graph structure with nodes as points and relationship strength as edge weights is constructed to generate a multi-monitoring point association map.

[0039] Preferably, the anomaly localization and strategy generation module specifically includes:

[0040] Risk score calculation is based on a multi-monitoring point correlation graph. Node paths are extracted, and the individual compliance status judgments and difference coefficient matrices corresponding to the paths are retrieved to calculate the node risk score.

[0041] Anomaly probability distribution is generated by statistically analyzing the risk scores of each route and combining them with route parameters to generate anomaly probability distribution maps.

[0042] Anomaly location involves performing spatial clustering on the anomaly probability distribution map to identify anomaly clusters and delineate anomaly boundaries.

[0043] A monitoring strategy is generated, which enables high-frequency monitoring in high-probability areas and applies test signals to adjacent points based on the anomaly probability distribution map.

[0044] Preferably, the anomaly location and strategy generation module further includes outputting a suspicious device identifier and an anomaly propagation path, wherein the suspicious device identifier is based on high-probability nodes in the anomaly probability distribution map, and the nodes are bound to actual devices to generate a list of suspicious devices;

[0045] Anomaly propagation paths are determined by simulating the anomaly diffusion process, recording path sequences, and identifying the most frequent path as the main propagation path.

[0046] Preferably, the test signal includes a multi-band injection signal, which is a frequency sweep signal injected through a controllable source;

[0047] The theoretical response is calculated based on the topological model;

[0048] The measured response was obtained through monitoring node data;

[0049] Anomaly detection is achieved by comparing the difference between the theoretical response and the measured response; when the difference exceeds a threshold, an anomaly is identified.

[0050] Preferably, the high-frequency monitoring includes increasing the sampling frequency, enabling real-time harmonic tracking, and deploying transient event detectors to record waveform segments.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] At the data processing level, the instrument transformer data acquisition and structuring module can effectively process the raw power data collected by the instrument transformer. It not only performs signal preprocessing of data such as current waveforms, voltage waveforms, and harmonic components, but also maps data to preset monitoring fields through field mapping and establishes inter-field association identifiers to generate standardized monitoring data units. This structured processing method makes the originally scattered raw data standardized and unified, effectively integrating data from different devices and monitoring points. This provides a unified data foundation for subsequent analysis work, breaking the traditional data processing paradigm of data fragmentation and difficulty in reuse.

[0053] In terms of theoretical value calculation, the theoretical monitoring value calculation module constructs a dynamic calculation model based on historical power data. Standardized monitoring data units are input into the model, and theoretical monitoring values ​​are output. Compared to traditional fixed models or static formulas, the dynamic calculation model can fully utilize the operational patterns inherent in historical data. It adjusts the calculation logic in real time as the power system's operating conditions change, making the output theoretical monitoring values ​​more closely match actual operating conditions. This allows for a more accurate reflection of the system's expected performance under different loads and operating states, providing a more reliable reference standard for subsequently judging whether measured data are normal.

[0054] The application of the rule compliance verification module changes the traditional, singular verification method. Based on standardized monitoring data units, this module accurately retrieves and matches operating rule clauses corresponding to different types of power equipment from the database, thereby generating a single compliance status judgment result. This approach of matching specific rules to different equipment types avoids the judgment bias caused by general rules, making compliance status judgments more targeted and accurate. It can more accurately identify situations where equipment does not comply with rules during operation, reduce the possibility of misjudgments and omissions, and promptly detect potential operational problems of the equipment.

[0055] The multidimensional difference analysis module enables a comprehensive comparative analysis of theoretical and measured monitoring values. It conducts analysis from multiple perspectives, including time-domain deviation, frequency-domain deviation, and waveform distortion, and generates a device-level difference coefficient matrix. This multidimensional analysis approach overcomes the limitations of traditional methods that only focus on numerical differences. It reveals the characteristics of the differences between theoretical and measured values ​​from different angles, more comprehensively uncovering potential anomalies in power system operation. This provides richer and more in-depth analytical basis for subsequent anomaly localization, helping to identify problems in system operation earlier and more accurately.

[0056] The network construction module constructs a graph structure with monitoring points as nodes and sharing relationships as edges by finding shared device identifiers and location information among different monitoring points, forming a multi-monitoring point association graph. This graph tightly links the originally independent data from each monitoring point, establishing a complete data association network. When an anomaly occurs in the system, the source of the anomaly can be quickly traced based on the association graph, and the scope of the anomaly's impact can be accurately determined, significantly improving the efficiency and accuracy of anomaly location and avoiding the difficulties in anomaly investigation caused by isolated data.

[0057] The anomaly localization and strategy generation module combines multi-monitoring point correlation maps, individual compliance status judgment results, and difference coefficient matrices to calculate risk scores and generate anomaly probability distribution maps. This allows for the localization of anomaly areas and the configuration of monitoring strategies. This systematic anomaly handling approach not only accurately locates anomalies but also develops targeted monitoring strategies based on anomaly probability distributions. Differentiated monitoring measures are implemented for anomalies of different regions and severity levels, ensuring timely and effective control of anomalies. This provides more comprehensive protection for the stable operation of the power system and avoids system operational risks caused by untimely or inappropriate anomaly handling. Attached Figure Description

[0058] Figure 1 This is a timing diagram of the intelligent power monitoring data analysis system based on current transformers described in this invention.

[0059] Figure 2 A flowchart illustrating the operation of the instrument transformer data acquisition and structured module;

[0060] Figure 3 A flowchart illustrating the operation of the rule compliance verification module. Detailed Implementation

[0061] 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, and 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.

[0062] Please see Figure 1 This invention provides an intelligent analysis system for power monitoring data based on instrument transformers. The system includes: an instrument transformer data acquisition and structuring module, a theoretical monitoring value calculation module, a rule compliance verification module, a multidimensional difference analysis module, an association network construction module, and an anomaly location and strategy generation module. These modules work together to achieve intelligent processing and analysis of power monitoring data.

[0063] The instrument transformer data acquisition and structuring module collects real-time power monitoring data from instrument transformers, including current waveforms, voltage waveforms, and harmonic components. This module performs signal preprocessing and field mapping operations, mapping the raw data to preset monitoring fields such as equipment number, current value, and voltage value. It also establishes inter-field association identifiers and generates standardized monitoring data units with a unified data format. The theoretical monitoring value calculation module constructs a dynamic calculation model using historical power data. This model receives standardized monitoring data units as input and outputs theoretical monitoring values ​​for subsequent comparison. The rule compliance verification module searches the database based on standardized monitoring data units, matches power equipment types with operating rule clauses, and generates a single compliance status judgment to assess equipment compliance. The multi-dimensional difference analysis module performs multi-dimensional difference analysis between theoretical and measured monitoring values, including time-domain deviation, frequency-domain deviation, and waveform distortion calculation, generating an equipment-level difference coefficient matrix to quantify the degree of deviation. The association network construction module, based on multiple standardized monitoring data units, searches for shared equipment identifiers and location information between different monitoring points, constructs a graph structure with monitoring points as nodes and shared relationships as edges, forming a multi-monitoring point association graph to reveal system correlations. The anomaly localization and strategy generation module is based on a multi-monitoring point correlation map. It combines single-item compliance status judgment and difference coefficient matrix to calculate risk scores, generates an anomaly probability distribution map to locate abnormal areas, and configures monitoring strategies such as adjusting sampling frequency or injecting test signals according to the distribution map to achieve intelligent anomaly response.

[0064] Example 1: See Figure 2The execution flow of the instrument transformer data acquisition and structuring module begins with the data acquisition unit. This unit captures raw current waveforms, voltage waveforms, and harmonic component data in real time through instrument transformers deployed on power equipment. It performs filtering operations to eliminate environmental noise and high-frequency interference, and uses normalization processing to scale raw signals of different dimensions to a uniform numerical range, generating a preprocessed data sequence to ensure the basic quality of the data. After receiving the preprocessed data sequence, the field mapping unit initiates a feature point recognition mechanism, scanning key feature points in the waveform sequence, including voltage peaks, current zero-crossing points, and harmonic amplitude abrupt change points. It establishes a mapping relationship between these feature points and preset monitoring fields. The preset fields cover core parameters such as the equipment's unique number, the effective value of the three-phase current, the instantaneous value of the line voltage, and the total harmonic distortion rate. The mapping process strictly follows the field definition dictionary to complete data annotation, generating a set of fields containing the raw monitoring values ​​as intermediate output. The structured processing unit performs format conversion operations on the field set, converting the time-series waveform data into a matrix storage format and reconstructing the field layout to optimize data access efficiency. For example, the current waveform is split into amplitude and phase sequences for independent storage, and harmonic components are represented using sparse matrix compression. Finally, a standardized mapping field set is generated to meet the requirements of structured storage. The association identification unit parses the logical relationships in the standardized mapping field set, automatically associates the corresponding location coordinates and topology information based on the device number, establishes inter-field association identifiers, and uses unique key-value pairs to achieve cross-table indexing. For example, association identifiers are bound to the current and voltage values ​​of the same device, and parent-child hierarchical identifiers are added to harmonic components and fundamental components. A standardized monitoring data unit encapsulating complete association relationships is generated and output to the downstream module.

[0065] This module employs a pipelined architecture to improve efficiency in real-time data processing. The data acquisition unit continuously outputs preprocessed data sequences at a fixed sampling rate. The field mapping unit accelerates feature point identification by processing multi-channel signals through parallel threads. The structured processing unit introduces memory pool management to reduce dynamic memory allocation overhead. The association identification unit utilizes a hash table to accelerate field association matching. Addressing the specific characteristics of power monitoring scenarios, the filtering operation uses an adaptive filter to dynamically adjust the cutoff frequency to cope with load fluctuations. Normalization processing retains the original data sign bits to avoid loss of effective information. Feature point identification incorporates a sliding window detection mechanism to prevent false triggering caused by transient interference. Field mapping uses dynamic template matching technology to be compatible with data formats from different manufacturers' devices. During structured conversion, a segmented compression storage strategy is used for high-frequency sampled waveform data to balance accuracy and storage cost. Field reconstruction follows power system data model specifications to ensure compatibility. When establishing association identification, the consistency between the device number and the topology database is automatically verified, and a real-time alarm mechanism is triggered for abnormal associations. The entire module handles high-concurrency data streams through a distributed message queue. Each standardized monitoring data unit carries metadata such as timestamps, device fingerprints, and data quality identifiers, providing directly callable structured input for subsequent analysis modules.

[0066] At the implementation level, the data acquisition unit employs a multi-stage filter bank design for filtering. The first stage performs power frequency bandpass filtering to retain the fundamental component; the second stage implements notch filtering to eliminate interference in specific frequency bands; and the third stage performs phase compensation to correct the transformer angle difference. Normalization processing synchronously records the scaling factor for data reconstruction. The feature point recognition algorithm of the field mapping unit includes three detection layers: the zero-crossing detection layer uses a sign comparison method to locate the waveform crossing zero; the peak detection layer determines the peak and trough positions through local extremum search; and the harmonic feature layer uses short-time Fourier transform to extract the amplitude of each harmonic. The mapping process completes data annotation through a predefined field rule engine. When performing binary conversion, the structured processing unit converts floating-point waveform data to fixed-point representation to improve processing speed. During the field reconstruction stage, related fields are merged to form data blocks to reduce I / O overhead; for example, current and voltage values ​​at the same timestamp are combined into a power calculation unit. The associated identification unit adopts a two-level association mechanism. The first level establishes field grouping based on the equipment number, and the second level builds electrical quantity association relationships within the group. Each data unit is marked by generating a globally unique identifier (UUID). The final output standardized monitoring data unit includes three parts: data body, association index, and quality label, forming a complete data package that can be directly used for database storage.

[0067] Multiple quality assurance measures are implemented during module operation. The data acquisition unit includes a signal amplitude rationality verification step, discarding abnormal data points exceeding the physically possible range. The field mapping unit initiates an interpolation compensation mechanism for unidentified feature points to avoid data loss. The structured processing unit ensures data conversion integrity through cyclic redundancy check (CRC). The association identification unit initiates a manual review process for conflict associations. For the transient process monitoring needs of the power system, the module is configured with a special event processing channel. When the waveform mutation rate exceeds a threshold, it automatically switches to high-speed processing mode, increasing the sampling rate and enabling a temporary storage buffer to ensure complete capture of fault waveform data. The standardized monitoring data unit adopts a hierarchical storage structure. Core monitoring fields are stored in the high-speed memory area, while waveform detail data is stored in the disk cache area. An intelligent preloading mechanism optimizes data access efficiency. This design ensures that the system can maintain millisecond-level response capability even in a thousand-node-level power grid monitoring scenario.

[0068] Example 2: See Figure 3The theoretical monitoring value calculation module initiates a historical data feature extraction process. This process decomposes the stored historical power monitoring data in multiple dimensions, uses a sliding window filtering algorithm to extract waveform segments under typical operating conditions, and extracts waveform feature parameters such as fundamental amplitude and phase shift. Harmonic analysis is performed using Fast Fourier Transform to establish a correlation mapping between the amplitude of each harmonic and the load type. A dynamic time warping algorithm is used to align waveform patterns at different time scales to eliminate the influence of sampling time difference. In the dynamic calculation model construction stage, the processed historical dataset is input into the prediction model framework. This framework includes a time-series prediction component based on equipment aging, uses a Long Short-Term Memory network to model the equipment performance degradation curve to generate basic predicted values, and a convolutional neural network with an embedded correction mechanism analyzes interference factors such as environmental temperature and humidity to correct measurement deviations. A real-time feature adjustment component dynamically updates the model weight parameters based on current operating data through an online learning mechanism. The theoretical value calculation operation parses the real-time input standardized monitoring data units into a format acceptable to the model, inputs them into the dynamic calculation model to perform forward propagation operations, and outputs a sequence of theoretical monitoring values ​​covering indicators such as current RMS value and voltage harmonic distortion.

[0069] The rule compliance verification module performs rule matching operations. This operation parses the equipment type code and operating status flag bits in the standardized monitoring data units, extracts field values ​​such as equipment rated parameters and real-time load rate, and performs pattern matching with the rule base. For missing fields, a strategy of using default values ​​of similar equipment is adopted to complete them, and a unified unit expression method is used, such as converting kilovolt-amperes to megavolt-amperes, forming a standardized set of field combinations. The rule complexity calculation process retrieves the set of matching rule clauses, analyzes the clause structure characteristics, including the number of nested conditions, the number of logical operators, the frequency of referencing external standards, the validity period span, and the geographical scope of application, and uses a weighted scoring model to calculate the rule complexity factor to quantify the execution difficulty. The compliance judgment stage selects a verification strategy based on the complexity factor. For low-complexity clauses, the monitoring data is directly called and the constraint threshold is compared numerically. For medium-complexity clauses, a multi-condition joint judgment is performed. For high-complexity clauses, the rule engine is launched to parse the logical expression, and condition threshold verification, numerical range verification, and timeliness judgment are completed item by item, generating a single compliance status judgment result with confidence score.

[0070] In the specific implementation of model construction, the time-series prediction component adopts a three-layer LSTM network structure. The input layer receives aging factors such as equipment commissioning time and cumulative operation count, the hidden layer models the nonlinear relationship of performance degradation, and the output layer generates the basic prediction of the theoretical current waveform. The correction neural network is designed with a dual-channel architecture. The environmental interference channel inputs temperature and humidity sensor data, and the electrical quantity channel receives voltage and current sample values. The deviation correction matrix is ​​output through the feature fusion layer. The real-time feature adjustment component implements an incremental learning mechanism, designing a sliding time window to cache the latest 200 sets of monitoring data. When the data distribution change within the window exceeds a threshold, model fine-tuning is triggered, and the parameters of the fully connected layer are updated using the stochastic gradient descent algorithm. In the rule matching stage, an automatic field completion algorithm is developed, and a device type-parameter mapping table is established to retrieve missing values. The unit conversion uses the International System of Units (SI) conversion coefficient matrix. The complexity calculation model defines a five-dimensional evaluation vector: nesting depth coefficient, logical operator density, reference dependency, time sensitivity coefficient, and regional weight. A complexity factor in the range of 0-1 is generated through normalized weighting. The compliance determination engine implements a multi-level processing pipeline: the primary validator handles simple threshold comparisons, the intermediate parser breaks down complex logical conditions, and the advanced executor calls the rule inference engine to handle nested reference clauses.

[0071] The data processing workflow implements multiple fault-tolerance mechanisms. An abnormal waveform filtering module is set up during the historical data feature extraction stage to automatically remove outlier data points exceeding ±3 standard deviations. Adversarial example enhancement technology is used in model training, injecting noisy synthetic data to improve robustness. A real-time performance guarantee module is embedded in the real-time calculation process, automatically switching to a simplified model output when theoretical value calculation times out. A conflict detection mechanism is configured in the rule matching stage, initiating priority arbitration when multiple conflicting rules are matched for the same field. Fuzzy logic is introduced to handle boundary cases in complexity factor calculation, avoiding rating jumps caused by minor parameter fluctuations. The compliance judgment output includes an additional evidence chain record, detailing the original data, cited clauses, and calculation paths involved in the judgment process. The system architecture adopts a microservice design, with the theoretical value calculation module deployed on a GPU-accelerated server cluster, and the rule verification module running on a distributed rule engine. Historical data is stored in shards through a time-series database, and a blue-green deployment strategy is used for model parameter updates to avoid service interruptions. Real-time data is distributed to computing nodes via a message queue, and a transactional confirmation mechanism is set up for result feedback. The rule base implements version management, supporting online hot updates of rule clauses without service interruption. The monitoring interface displays a real-time comparison curve of theoretical and measured values. The compliance status is presented intuitively through traffic light indicators. The complexity factor is visualized in the form of a heat map to show the load distribution of the rule system.

[0072] To address the dynamic characteristics of power systems, the model incorporates a unique adaptive mechanism: when a short-circuit fault transient process is detected, the aging factor update is automatically frozen to prevent mislearning; during periods of severe load fluctuations, the sampling window of the real-time feature adjustment component is shortened to the 10-second level; and dedicated rule templates are configured for renewable energy access nodes, with added special verification clauses for harmonic distortion rate. The output adopts a hierarchical structure: the base layer contains theoretical values ​​and compliance status indicators; the enhancement layer adds confidence scores and complexity ratings; and the audit layer retains complete calculation logs and rule reference chains to meet the needs of different application scenarios.

[0073] Example 3: After receiving the theoretical monitoring values ​​and the measured monitoring data units, the multidimensional difference analysis module initiates the time-domain deviation calculation process. This process performs a sliding comparison operation on the two sets of data sequences with a 500-millisecond time window. A dynamic time warping algorithm is used to align the time axes of the theoretical and measured values ​​to eliminate sampling delay differences. The cumulative absolute deviation of the data points within the calculation window generates a time-domain deviation vector to record the time dimension offset. The frequency-domain deviation detection process performs a Fast Fourier Transform to convert the theoretical harmonic components and the actual harmonic distribution to the frequency domain. The energy spectral density ratio of each harmonic is calculated to reflect the energy distribution difference. The deviation indices of the 3rd, 5th, and 7th characteristic harmonics are extracted to construct a frequency-domain deviation vector to quantify the spectral characteristic changes. The waveform distortion assessment module initiates a dynamic morphological matching algorithm to calculate the segment-by-segment similarity between the theoretical waveform template and the measured waveform sequence. Curvature change analysis quantifies the degree of waveform distortion, generating a distortion vector to characterize morphological anomalies. The difference coefficient matrix generation process concatenates three vectors into a composite vector and uses a weighted fusion strategy to integrate multi-dimensional features. The time domain weight is set to 0.4, the frequency domain weight to 0.3, and the distortion weight to 0.3. After normalization, a standardized difference coefficient matrix is ​​output. The rows of this matrix correspond to the monitoring equipment, and the columns represent the deviation dimensions.

[0074] The association network construction module first performs a shared entity extraction operation, scanning the device identification field and GPS coordinate field of multiple standardized monitoring data units. It then uses a fast matching algorithm based on prefix trees to identify identical device numbers or adjacent location codes. For duplicate entities, a Bloom filter is applied to remove duplicates and generate a shared entity set. The association density calculation process statistically analyzes the frequency and spatial distribution density of each entity in the monitoring point set, and designs a density index calculation model.

[0075]

[0076] in: This represents the density index between monitoring points i and j. and Record the number of times the entity appears at each of the two points. The distance between the two points is the Euclidean distance (in kilometers). During the graph structure construction phase, monitoring points are encoded as graph nodes, with the density index used as the edge weight. A bidirectional connection edge is established, and an adjacency list storage structure is used to generate a multi-monitoring point association map.

[0077] The time-domain deviation calculation employs a three-level processing architecture. The first level performs data resampling to align the theoretical and measured values ​​with a time reference. The second level applies a dynamic time warping algorithm to calculate the minimum path cumulative deviation. The third level smooths random fluctuations through sliding window mean filtering to generate a stable deviation vector. Frequency domain analysis includes a special processing channel, adding an interharmonic component detection channel for new energy access nodes and configuring special harmonic weighting coefficients for electric arc furnace load nodes. A multi-scale template library is developed for waveform matching algorithms, including three benchmark forms: steady-state waveform templates, transient impact templates, and resonant oscillation templates. The matching process automatically selects the optimal template to calculate the structural similarity index. The weighting coefficients of the difference matrix are dynamically adjusted according to the equipment type, increasing the frequency domain weight to 0.5 for transformer equipment and emphasizing the time domain weight to 0.6 for transmission line equipment.

[0078] The shared entity extraction implements spatial index optimization, establishes an R-tree spatial database to accelerate geographic location retrieval, and uses a distributed hash table for device identifier matching to improve query efficiency. The density index calculation introduces a decay factor to process historical data, assigning a time decay coefficient of 0.7 to data older than 24 hours. The graph construction process employs a layered design: the bottom-level topology graph records physical connections, while the upper-level association graph overlays electrical coupling relationships, setting minimum connectivity thresholds for important hub nodes to force the retention of critical edges. Graph storage is implemented using the graph database Neo4j; node attributes include metadata such as device type and rated parameters, while edge attributes store the density index and the most recent update timestamp. The system operation implements a real-time guarantee mechanism: time-domain analysis uses a streaming computing framework to update the deviation vector every 200 milliseconds, and frequency-domain analysis utilizes an FPGA acceleration card to improve FFT calculation speed. The waveform matching module is configured with a dedicated memory pool to cache template library data, and the difference matrix generation uses a double buffer to achieve lock-free updates. The association network construction adopts an incremental update strategy; when new monitoring data is added, only the affected subgraph is locally updated, significantly reducing computational overhead. The visualization interface renders deviation heatmaps and network topology graphs in real time, supporting a 3D spatial perspective to view anomaly propagation paths.

[0079] Adaptive processing is designed for special scenarios in power systems: during lightning overvoltage events, the time-domain deviation threshold is automatically relaxed to avoid false alarms; frequency-domain monitoring channels are added for nodes in resonant regions; and higher density alarm thresholds are set for associated nodes of older equipment. Data output uses binary protocol encapsulation, the difference coefficient matrix is ​​transmitted in single-precision floating-point array format, and the correlation graph is exported in the GraphML standard format to support parsing by third-party analysis tools. The monitoring backend can view the three-dimensional deviation vector component change curves of any node in real time, and the correlation graph supports force-directed layout to dynamically display the evolution of the network structure. In terms of computing resource management, the time-domain analysis thread is bound to a high-performance CPU core to ensure real-time performance, the frequency-domain analysis task is scheduled to a processor with the AVX instruction set, and waveform matching uses GPU parallel computing to accelerate template comparison. Memory management adopts object pool reuse technology to avoid frequent memory allocation and release. The exception handling mechanism includes a data verification module, which automatically triggers a data retransmission mechanism when an abnormal jump in theoretical value is detected, and starts a local cache recovery mode when the correlation network update fails. The communication interface uses zero-copy technology to reduce data transmission latency, and cyclic redundancy check codes are added to important data frames to ensure integrity. The difference coefficient matrix is ​​designed with scalability in its row and column dimensions, automatically expanding the column vectors when a new monitoring dimension is added and dynamically adding row records when a new device node is added. The matrix storage uses a sparse matrix compression format, omitting deviation values ​​below a threshold to save space. The association graph implements a dynamic pruning strategy, periodically removing weak connections with a density index below 0.3 to maintain a simple network structure. The system maintenance interface supports injecting simulated test data to verify the analysis process and allows exporting deviation matrices and graph snapshots for any time period for post-analysis.

[0080] Example 4: The anomaly location and strategy generation module initiates the risk score calculation process. This process extracts the three-layer node path centered on the substation busbar from the multi-monitoring point association map, retrieves the individual compliance status judgment results and difference coefficient matrix data corresponding to each node on the path, assigns a weight coefficient of 0.7 to compliance status violations, and assigns a weight coefficient of 0.9 to difference matrix exceedances, and calculates the node risk score value through weighted summation. The anomaly probability distribution generation operation statistically analyzes the risk scores of all nodes at the feeder level, combines parameters such as line load rate and equipment commissioning years for normalization processing, and generates a two-dimensional gridded anomaly probability distribution map, where color depth represents the risk level. Anomaly area location executes a density-based spatial clustering algorithm to identify continuous areas with probability values ​​greater than 0.85, automatically delineating irregular polygonal boundaries to mark anomaly areas. The monitoring strategy generation unit configures response measures according to the distribution map, enabling a high-frequency sampling mode of 5000 points per second for grid areas with probability values ​​exceeding 0.9, and applying a 10V / 50Hz test signal to adjacent grid nodes.

[0081] The module synchronously outputs a list of suspicious device identifiers. This list parses high-probability nodes in the anomaly probability distribution map and binds the node IDs to the actual physical devices through a device code mapping table, generating a list of suspicious devices sorted in descending order of risk score. Anomaly propagation path analysis initiates a simulation based on a graph neural network, injecting anomaly signal sources into the correlation graph, recording the signal propagation path sequence, and counting the activation frequency of each path. The path with the highest frequency is determined as the main propagation path (see Table 1).

[0082] Table 1: Output Table of Anomaly Location Module

[0083]

[0084] In a specific implementation case, the monitoring system of a 110kV substation detected abnormal bus voltage fluctuations. Risk scoring calculation first located the critical path in the correlation graph: main transformer monitoring point (B-1101) → outgoing switch (S-2056) → feeder terminal (T-3089). Searching the B-1101 node showed its voltage compliance status as "operating above the upper limit," with a time-domain deviation value of 0.35 in the difference matrix; the waveform distortion difference coefficient for the S-2056 node was 0.41; and the temperature compliance status for the T-3089 node was "warning." Weighted calculation yielded a score of 0.92 for B-1101, 0.87 for S-2056, and 0.78 for T-3089. The anomaly probability distribution map showed a high-risk area centered on B-1101 (probability > 0.9), and spatial clustering delineated an elliptical anomaly area covering the three nodes. The strategy generation unit activates the transient event detector to record waveform segments for B-1101 and injects a 50-150Hz sweep frequency test signal into S-2056. The suspicious device identification output maps node B-1101 to main transformer #1 (model SZ11-50000 / 110) and S-2056 to switchgear 2056, adding them to the priority investigation list. The anomaly propagation simulation shows that the anomaly spreads from main transformer #1 through switchgear 2056 to the feeder, with the main propagation path marked as "main transformer medium voltage side → switchgear busbar room → feeder CT secondary circuit". During test signal injection, a controllable source applies a 10V / 50Hz reference signal superimposed with a 2V / 150Hz harmonic signal at the S-2056 monitoring point. The theoretical response model predicts that an 8.5V / 50Hz + 1.8V / 150Hz signal should be detected at the adjacent T-3089 point. Actual response data shows that the T-3089 only captured the 6.2V / 50Hz fundamental component, with the 150Hz component missing. The difference distance exceeded the threshold, triggering a "secondary circuit open circuit" anomaly flag. High-frequency monitoring implementation details included: increasing the sampling frequency of monitoring point B-1101 to 5000 points / second; enabling real-time harmonic tracking mode to update the harmonic spectrum every 100 milliseconds; deploying a transient event detector with a 20ms pre-trigger buffer to record the waveform before the sudden change. The test signal was generated using a programmable signal source with an output impedance matched to the 600Ω power communication standard; the signal injection point was selected at the switchgear test terminal block. Response analysis used a 10ms time window to synchronously compare the theoretical response with the measured waveform; the difference distance calculation employed a dynamic time warping algorithm to eliminate the effect of time delay.

[0085] The system architecture employs a distributed deployment, with the risk scoring calculation engine running on a real-time analysis server and anomaly probability distribution generation utilizing GPU parallel rendering. The spatial clustering algorithm is optimized for multi-threading, processing 1000 nodes in under 50 milliseconds. Monitoring strategy configurations are distributed to field monitoring devices via the standard Modbus protocol, with strategy switching response time controlled within 200 milliseconds. The suspicious device list output integrates with the enterprise asset management system interface, automatically generating maintenance work orders. Propagation path visualization utilizes WebGL technology for 3D dynamic display, supporting overlay display of topology maps and geographic information. A quality monitoring loop is implemented in data processing, recording the quality indicators of each input parameter during risk scoring calculation and automatically downgrading low-quality data. Anomaly area boundary generation is optimized using a convex hull algorithm to avoid including normal nodes in anomaly areas. Security checks are performed before test signal injection to confirm the system has no risk of protective actions. High-frequency monitoring data is stored in segments with compression, and time synchronization markers are added to data packets every 5 minutes. The system is designed for environmental adaptability: in areas with severe electromagnetic interference, the test signal amplitude is increased to 15V to ensure a high signal-to-noise ratio; and the high-frequency monitoring data retention period is extended for equipment in humid environments. The output report includes an anomaly area location map, a list of suspicious devices, and a propagation path analysis. Figure 3 The system supports exporting to PDF and SCADA system standard formats. The system maintenance interface allows setting risk threshold parameters and supports importing power grid structure change information for real-time updates to the associated graph. In emergency scenarios involving communication interruptions, the local monitoring device continues high-frequency data acquisition according to the last received policy instructions, with cached data to be retransmitted after communication is restored. Important operations, such as test signal injection, require security authentication from the main station to prevent accidental operation. The historical data tracing function supports replaying the evolution of anomaly probability distributions for any time period, assisting in the analysis of fault development trajectories.

[0086] Example 5: The test signal implementation process in the anomaly location and strategy generation module is initiated when an abnormal waveform distortion is detected at the switchgear monitoring point S-2056. A controllable signal source outputs a multi-band injection signal containing a 10V / 50Hz fundamental component superimposed with a 2V / 150Hz harmonic component, sweeping the frequency range from 50-250Hz to excite the system's frequency response characteristics. The theoretical response calculation is based on the power grid topology model, extracting the line parameters (length 1.2km, cross-sectional area 240mm², copper core) from S-2056 to the adjacent monitoring point T-3089. Considering distributed capacitance and proximity effects, the signal propagation attenuation characteristics are deduced using transmission line equations, predicting a standard response waveform of 8.5V / 50Hz + 1.8V / 150Hz at point T-3089. The measured response acquisition is performed through the data acquisition unit of the T-3089 monitoring node, using synchronous sampling technology to capture the response signal with a 1μs time accuracy. The actual waveform data packet includes timestamps, amplitude sequences, and spectral analysis results. The anomaly detection process dynamically aligns the theoretical response waveform with the measured data using time warping. The improved Frescher distance algorithm is used to calculate the waveform shape difference distance. When the difference distance of the 150Hz component exceeds the preset threshold of 0.35, the system automatically marks "secondary circuit anomaly" and triggers alarm code E207.

[0087] The high-frequency monitoring strategy is activated at the B-1101 main transformer monitoring point where the risk score exceeds the limit. The sampling frequency is increased from the conventional 1kHz to 5000 points / second, and the analog-to-digital converter adopts a Σ-Δ architecture to ensure high-resolution sampling. The harmonic real-time tracking function uses a 128-point sliding window fast Fourier transform, updating the harmonic spectrum distribution map every 100 milliseconds, focusing on monitoring the amplitude change rate of the 3rd to 13th harmonics. The transient event detector is configured with a ring buffer to achieve 20ms pre-trigger storage. When a waveform change rate exceeding 50V / ms is detected, it automatically saves the waveform segments before and after 100ms, and the segment data is marked with GPS time synchronization. The test signal generation is implemented using a programmable function generator, and the output stage is configured with a power amplifier to provide a maximum drive capability of 20V / 5A. The output impedance is adjusted to 600Ω through an automatic matching network. The signal injection point is selected from the voltage monitoring loop of the switch cabinet test terminal block, and strong crosstalk is prevented through an opto-isolation coupler. The response acquisition system uses a 24-bit high-precision ADC chip, the anti-aliasing filter is set with a cutoff frequency of 150kHz, and the sampling clock is strictly synchronized with the injected signal source through a PLL circuit. A dedicated coprocessor is deployed for difference distance calculation, performing waveform similarity comparison every 10 milliseconds, with the results written to the anomaly detection register in real time. The high-frequency sampling channel's analog front-end circuitry is reconfigured, the gain amplifier is switched to low-noise mode, and the anti-aliasing filter cutoff frequency is increased to 25kHz. The harmonic tracking algorithm employs a parallel processing architecture, with the fundamental and harmonic channels processed separately, and the spectrum update results transferred to shared memory via DMA. The transient detector has multi-level trigger conditions: Level 1 triggering is based on amplitude mutation detection (>15% of rated value), Level 2 triggering is based on waveform derivative threshold (>30V / ms), and Level 3 triggering is based on harmonic distortion rate mutation (>5% change). The system implements multiple safety mechanisms, automatically detecting the equipment's insulation status before test signal injection, and blocking signal output when the insulation resistance is below 2MΩ. Data integrity protection is activated during high-frequency monitoring, with a CRC-32 checksum added to each frame of sampled data. Anomaly marking uses a progressive confirmation strategy: the first exceedance of the threshold triggers a warning state, and three consecutive exceedances convert to a confirmed state. The communication protocol adopts the industrial-grade Modbus / TCP standard, with key commands including dual authentication.

[0088] In a specific case, a 220kV substation detected a risk score of 0.94 at monitoring point B-2103 on the medium-voltage side of main transformer #2. The system automatically issued instructions: to enable a sampling rate of 5000 points / second for B-2103, activate the real-time harmonic tracking mode, and deploy a transient event detector; to inject a test signal (12V / 50Hz + 3V / 250Hz) into the adjacent switchgear S-3157. Theoretical response model calculations showed that a signal of 10.2V / 50Hz + 2.1V / 250Hz should appear at downstream point T-4201. Actual measured data captured an actual signal of 9.8V / 50Hz + 0.3V / 250Hz at point T-4201, with a 250Hz component difference distance of 0.41. The system determined "abnormal high-frequency response of the CT secondary circuit," and the transient detector recorded three voltage dip events with a duration of 2ms. The signal injection device employs a modular hardware design. The fundamental frequency generator utilizes direct digital frequency synthesis technology, while harmonic components are generated through an FIR filter bank. The output stage is equipped with an overcurrent protection circuit to limit the maximum output current. The response acquisition unit is equipped with an independent power supply and uses fiber optic isolation to transmit sampled data. High-frequency monitoring data streams undergo graded compression: the fundamental frequency component retains its original accuracy, harmonic components are lossy compressed, and transient waveform segments are stored completely in floating-point format. The operation monitoring interface displays the test signal spectrum, theoretical / measured response comparison curves, and the trend of difference distance changes in real time. The system supports historical data backtracking analysis, allowing access to anomaly judgment process records for any time period. The maintenance toolkit provides signal injection calibration functionality and supports importing standard tables to compare data and verify measurement accuracy. Adaptive handling for special operating conditions is included: automatically reducing the test signal amplitude to a safe threshold during the rainy season; limiting high-frequency sampling time at monitoring points on older equipment to prevent overheating; and disabling specific frequency test signals in system resonance risk areas. The data output interface supports the IEC61850 protocol, and anomaly marker information is uploaded to the dispatch master station in real time. The equipment's self-test procedure performs signal loop continuity tests daily and periodically calibrates the sampling channel gain error. A complete operation log is recorded during implementation, including signal injection parameters, theoretical response calculation basis, measured data fingerprints, and key parameters of the judgment process. All operation commands are digitally signed for authentication, and data storage is encrypted using AES-256. The system supports remote diagnostic mode, allowing experts to adjust anomaly judgment thresholds online or manually trigger specific test sequences.

[0089] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0090] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent analysis system for power monitoring data based on current transformers, characterized in that, include: Instrument transformer data acquisition and structuring module: Based on the real-time acquisition of power monitoring data from instrument transformer equipment, including current waveform, voltage waveform and harmonic components, the module performs signal preprocessing and field mapping, maps the data to preset monitoring fields, establishes inter-field association identifiers, and generates standardized monitoring data units. Theoretical monitoring value calculation module: Based on historical power data, a dynamic calculation model is constructed. The standardized monitoring data unit is input into the dynamic calculation model, and the theoretical monitoring value is output. Rule compliance verification module: Based on the standardized monitoring data unit, it retrieves data from the database to match the type of power equipment with the operating rule clauses, and generates a single compliance status judgment; Multidimensional difference analysis module: Performs multidimensional difference analysis on the theoretical monitoring value and the measured monitoring value. The multidimensional difference analysis includes time domain deviation, frequency domain deviation and waveform distortion, and generates a device-level difference coefficient matrix. The associated network construction module: Based on multiple standardized monitoring data units, it searches for shared device identifiers and location information among different monitoring points, constructs a graph structure with monitoring points as nodes and shared relationships as edges, and establishes a multi-monitoring point association graph; Anomaly location and strategy generation module: Based on the multi-monitoring point association map, combined with the single compliance status judgment and the difference coefficient matrix, calculate the risk score, generate an anomaly probability distribution map, locate the anomaly area, and configure the monitoring strategy according to the anomaly probability distribution map; The theoretical monitoring value calculation module specifically includes: Historical data feature extraction involves multi-dimensional decomposition of historical power monitoring data, including using filtering algorithms to extract waveform features, establishing the correlation between harmonics and load through harmonic analysis, and aligning waveform patterns using sequence analysis algorithms. The dynamic calculation model is constructed by inputting the processed historical data into the prediction model. The prediction model includes a time-series prediction component based on equipment aging to generate basic prediction values, a neural network with embedded correction mechanisms to correct measurement deviations, and a real-time feature adjustment component to dynamically adjust the prediction based on real-time data. Theoretical value calculation involves inputting real-time standardized monitoring data units into the dynamic calculation model to obtain theoretical monitoring values; The rule compliance verification module specifically includes: Rule matching is based on the equipment type field and operating status field in the standardized monitoring data unit. The field values ​​are extracted and matched with the rule base to fill in missing fields, unify the expression method, and obtain a standardized set of field combinations. The rule complexity calculation involves retrieving matching rule clauses based on a set of normalized field combinations, extracting the clause's nesting level, number of logical judgments, number of references, time range, and geographical coverage, and then calculating the rule complexity factor. The compliance determination is based on the rule complexity factor. It calls the monitoring data and clause constraint values ​​item by item, performs condition threshold comparison, numerical range verification and timeliness judgment, and generates a single compliance status determination. The anomaly localization and strategy generation module specifically includes: Risk score calculation is based on a multi-monitoring point correlation graph. Node paths are extracted, and the individual compliance status judgments and difference coefficient matrices corresponding to the paths are retrieved to calculate the node risk score. Anomaly probability distribution is generated by statistically analyzing the risk scores of each route and combining them with route parameters to generate anomaly probability distribution maps. Anomaly location involves performing spatial clustering on the anomaly probability distribution map to identify anomaly clusters and delineate anomaly boundaries. A monitoring strategy is generated, which enables high-frequency monitoring in high-probability areas and applies test signals to adjacent points based on the anomaly probability distribution map.

2. The intelligent analysis system for power monitoring data based on instrument transformers according to claim 1, characterized in that, The current transformer data acquisition and structuring module specifically includes: The data acquisition unit collects raw power data through the current transformer, performs filtering and normalization processing, and generates a preprocessed data sequence. The field mapping unit identifies feature points in the preprocessed data sequence and maps them to preset monitoring fields, including device number, current value, voltage value, and harmonic content, to generate an original field set. The structured processing unit performs format conversion and field reconstruction on the original field set to generate a standardized mapping field set; The associated identification unit performs internal association matching on fields in the standardized mapping field set, establishes association relationship identifiers, and generates standardized monitoring data units.

3. The intelligent analysis system for power monitoring data based on current transformers according to claim 2, characterized in that, The multidimensional difference analysis module specifically includes: The time-domain deviation calculation involves sliding comparison of theoretical and measured monitoring values ​​using a time window, aligning sequences using a sequence alignment algorithm, calculating cumulative deviation, and generating a time-domain deviation vector. Frequency domain deviation detection involves performing frequency domain analysis on the harmonic components of theoretical and measured values, calculating the energy spectral density ratio, extracting the harmonic deviation index, and constructing a frequency domain deviation vector. Waveform distortion assessment is based on a matching algorithm that compares the theoretical values ​​with the morphology of the measured waveform sequences, calculates the degree of waveform distortion, and generates a distortion vector. The difference coefficient matrix is ​​generated by concatenating the time-domain deviation vector, frequency-domain deviation vector, and distortion vector, and then outputting the difference coefficient matrix through weighted normalization.

4. The intelligent analysis system for power monitoring data based on current transformers according to claim 3, characterized in that, The associated network construction module specifically includes: Shared entity extraction is based on multiple standardized monitoring data units. The equipment identification field and location field in each unit are extracted, and entity classification, matching and deduplication are performed to generate a shared entity set. The correlation density calculation involves statistically analyzing the frequency and category of shared entities at different monitoring points, recording distribution information, and calculating the correlation density index. Graph structure construction: Based on the association density index, the monitoring points are coded into nodes, the shared relationship strength value between nodes is calculated, and a graph structure with nodes as points and relationship strength as edge weights is constructed to generate a multi-monitoring point association map.

5. The intelligent analysis system for power monitoring data based on current transformers according to claim 4, characterized in that, The anomaly location and strategy generation module also includes outputs including suspicious device identifiers and anomaly propagation paths. The suspicious device identifiers are based on high-probability nodes in the anomaly probability distribution map, which are then bound to actual devices to generate a list of suspicious devices. Anomaly propagation paths are determined by simulating the anomaly diffusion process, recording path sequences, and identifying the most frequent path as the main propagation path.

6. The intelligent analysis system for power monitoring data based on current transformers according to claim 5, characterized in that, The test signal includes a multi-band injection signal, which is a frequency sweep signal injected through a controllable source; The theoretical response is calculated based on the topological model; The measured response was obtained through monitoring node data; Anomaly detection is achieved by comparing the difference between the theoretical response and the measured response; when the difference exceeds a threshold, an anomaly is identified.

7. The intelligent analysis system for power monitoring data based on current transformers according to claim 6, characterized in that, The high-frequency monitoring includes increasing the sampling frequency, enabling real-time harmonic tracking, and deploying transient event detectors to record waveform segments.

Images