Electric quantity acquisition terminal adaptive fault studying and judging system based on grey relational degree analysis

Abstract

The invention discloses an electric quantity acquisition terminal self-adaptive fault studying and judging system based on grey correlation degree analysis, and particularly relates to the technical field of electric power system operation state monitoring, which comprises five function modules of acquisition, standardization, identification, judgment and strategy generation, an operation data construction time sequence is acquired, a grey correlation coefficient sequence is formed through standardization, and an electric quantity acquisition terminal self-adaptive fault studying and judging system is established. The association factors lower than a threshold value are recognized to generate an abnormal feature set, fault types are recognized through map matching, priority ranking is executed, finally, a response process is generated, related channel collection periods are dynamically adjusted, and accurate recognition of the fault types and self-adaptive control over collection behaviors are achieved; according to the invention, dynamic comparison between multi-channel data and a reference state is realized, and the fault identification accuracy is enhanced; through structural extraction of an abnormal feature set and execution of label matching, the complex fault discrimination capability is improved; a response strategy is automatically generated in combination with the fault type, the sampling period is dynamically adjusted, and the system adaptability and the resource utilization efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of power system operation status monitoring technology, and more specifically, to an adaptive fault assessment system for power acquisition terminals based on grey relational analysis. Background Technology

[0002] With the continuous construction and advancement of smart grids and digital distribution systems, power acquisition terminals, as the main devices for acquiring basic power information, undertake the functions of real-time acquisition, preliminary processing, and network transmission of various electrical parameters such as voltage, current, active power, reactive power, and frequency. They are important front-end devices supporting power grid metering, anomaly diagnosis, energy efficiency assessment, and condition monitoring. However, the actual operating environment is complex and variable. Power acquisition terminals are often affected by factors such as non-standard wiring, load disturbances, communication anomalies, power fluctuations, and electromagnetic interference, leading to abnormal behaviors such as channel signal drift, abrupt changes, and interruptions. This, in turn, causes a series of problems such as data quality degradation, expansion of monitoring blind spots, and strategy execution failure.

[0003] Existing technologies for monitoring anomalies in data acquisition terminals mostly employ static rule matching, empirical threshold judgment, or single-channel anomaly triggering. These methods often rely on manually set static models, making it difficult to adapt to the complex evolution of correlations between electrical parameters in dynamic scenarios. Especially when dealing with multi-factor coupling, periodic fluctuations, and nonlinear interference, they are prone to misjudgments and missed detections, hindering accurate identification and rapid response. Furthermore, current terminal acquisition cycles are mostly statically configured. Even with significant fluctuations in the power channel, the system continues sampling at a fixed frequency, failing to perform high-frequency tracking of short-term anomalies or release sampling cycles for stable channels, thus creating a structural contradiction between sampling resource utilization efficiency and data validity. Therefore, this invention proposes an adaptive fault assessment system for power acquisition terminals based on grey relational analysis to address the aforementioned problems. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: An adaptive fault assessment system for power acquisition terminals based on grey relational analysis includes a data acquisition and processing module, a sequence standardization module, an anomaly factor identification module, a fault type determination module, and a response strategy generation module. The data acquisition and processing module collects operational data from multiple power channels, constructs corresponding time series for each channel, and processes the data. At the same time, it calls the stored normal operation standard sequence as a reference object, i.e., the reference sequence. The sequence normalization module performs numerical normalization on all processed running data sequences and reference sequences, and calculates the grey relational coefficient at each corresponding sampling point to form a relational coefficient sequence, which is used to reflect the instantaneous similarity of each sampling channel at each time point. The abnormal factor identification module averages the grey relational coefficient sequence over time to obtain the full sequence correlation value for each power channel. When the correlation of a certain item is lower than its corresponding set threshold, it extracts the low correlation factor combination for the current period by combining its corresponding channel information and power parameter type. It then summarizes all low correlation factor combinations to form the correlation abnormal feature set for the current period. The fault type determination module is based on the pre-built mapping relationship between abnormal features and fault labels. It uses the generated feature set to perform graph matching in the fault label space to determine the specific fault type, and determines whether there are multi-factor composite anomalies, and sorts and classifies the types according to their priority. The response strategy generation module retrieves the corresponding response process from the policy rule base based on the identified fault type, writes the operation instructions in the response process into the execution buffer, and dynamically adjusts the collection cycle of the corresponding power channel.

[0005] In a preferred embodiment, data processing refers to: Time series are synchronized, missing data are filled in, and numerical smoothing is performed to form a sequence structure of equal length and without abrupt changes.

[0006] In a preferred embodiment, the starting point zeroing method is used during numerical normalization, and the operating data sequence after the power channel processing is defined as follows: ; Define the reference sequence as: m and n are both initial index numbers and have the same value; After processing using the initial point zeroing method, we obtain: ; ; in, It can be deduced that: ;at this time, The value can be 0 or 1. This is the index of the number after the initial point zeroing method.

[0007] In a preferred embodiment, calculating the grey relational coefficient at each corresponding sampling point refers to: Within a fixed time window, assume the sequence Given an n-1 time series, then the sequence... In the interval The corresponding graph above should be a broken line. ,in , ; sequence Corresponding broken line In the interval The change in area can be expressed as: ; For sequence and In the interval The change in area on is denoted as and ; but ; For sequence and The grey relational coefficient, at this point, and They are 0 and 1 respectively.

[0008] In a preferred embodiment, during the fault type determination process, the associated abnormal feature set identifies fault type labels through a graph matching method. The graph matching process includes performing deviation score calculation for each preset label and filtering out all labels with scores lower than a preset identification threshold based on the score results to form the identifiable label set for the current period. The identifiable tag set is used to represent all candidate fault types that meet the set requirements for matching abnormal features within the current period. If the set contains only one tag, then that tag is determined as the final fault type tag for the current period. If the set contains multiple tags, then no type decision is made for the time being, and the subsequent priority sorting and classification process begins.

[0009] In a preferred embodiment, the deviation scoring calculation mechanism includes two consecutive calculation stages: the first stage is matching position identification, and the second stage is score calculation. In the first stage, for each low-correlation factor combination contained in the associated abnormal feature set generated in the current cycle, its index position in the two-dimensional structure composed of the power channel and the combination of power parameter types is extracted to form the index sequence of the feature of this cycle, which serves as the basis for spatial comparison. Then, the index sequence is compared item by item with the label index sequence constructed in the initialization stage for each fault type label in the fault label space. The number of positions with completely identical indices is counted, and this number is used as the baseline for the matching degree score between the feature of this cycle and the corresponding label. In the second stage, the correlation values ​​of the current period corresponding to all factors with consistent index positions are extracted, and the deviation is calculated with the average correlation value of the training samples corresponding to the corresponding position of each label in the label index sequence. Each deviation value is weighted according to the factor importance weight set in the label space construction stage. Finally, all weighted deviation values ​​are summed to form a complete deviation score. The above calculation is repeated for all labels, and all labels with deviation scores lower than the recognition threshold are selected to form the set of identifiable labels for the current period.

[0010] In a preferred embodiment, when the identifiable tag set contains multiple tags, a type priority sorting and classification operation is performed. During the sorting process, the priority parameter corresponding to each tag in the identifiable tag set is read and arranged from high to low according to the parameter value. The tag with the first position in the sorting is determined as the dominant fault type in this cycle, and the remaining tags are classified into secondary response types in the sorting order. The priority parameter used for sorting comes from the response importance level value set in the tag space initialization stage. This value is used to classify and evaluate the impact of historical response delays on the system. After sorting, only the response process corresponding to the dominant fault type is used for control logic generation and written into the execution buffer.

[0011] In a preferred embodiment, during the response strategy generation process, the acquisition instructions contained in the execution buffer undergo pre-scheduling processing. Based on the current cycle correlation degree value corresponding to each low correlation factor combination in the current cycle correlation anomaly feature set, the acquisition cycle of the mapped power channel is dynamically adjusted. The adjustment method is as follows: For each acquisition command, check if there is a matching item in the associated abnormal feature set for its target channel. If there is, extract the current cycle correlation value corresponding to the electrical parameter combination of the channel and compare the value with the preset cycle adjustment judgment range. If the value is below the lower limit range, shorten the acquisition cycle of the channel to a certain proportion of the original cycle to enhance the time capture frequency of abnormal fluctuation signals. If the value is below the upper limit but does not exceed the lower limit, the original collection cycle remains unchanged to maintain normal collection status. If the value is above the normal range, the acquisition period is multiplied by a preset extension coefficient to obtain the adjusted acquisition period, which is used to release data channel load resources and reduce storage overhead. All period adjustment decisions are completed before the execution buffer scheduling and are embedded in the response process as part of the power acquisition and control behavior of this period, forming a closed-loop response structure.

[0012] The technical effects and advantages of this invention are as follows: This invention collects operational data from multiple power channels through a data acquisition and processing module, constructs time series for each channel, and compares them with stored reference sequences, establishing a structured comparison system between sampled data from different channels and standard operating states. A sequence standardization module normalizes the sampled data and calculates a grey relational coefficient, reflecting the instantaneous similarity between the sampled channel and the reference state at each time point. Then, an anomaly factor identification module extracts the correlation along the time dimension, ensuring that fault determination focuses not only on individual anomalies but also on trend analysis based on the entire cycle. This approach improves the stability of anomaly identification, reduces the risk of misjudgment due to instantaneous fluctuations, and effectively enhances the system's overall perception of the power channel operating status.

[0013] This invention supports structured anomaly feature extraction based on grey relational analysis, enhancing the system's ability to judge complex faults. By averaging the grey relational coefficient sequence to obtain the full-sequence relational value for each channel, and combining channel information with electrical parameter types to form low-correlation factor combinations, a set of correlated anomaly features is formed within a single acquisition cycle, achieving a structured expression of anomaly factors. This anomaly aggregation method based on channel parameter pairs not only retains the dual characteristics of fault location and type but also establishes a multi-dimensional anomaly information integration pathway. The fault type determination module matches this feature set with a preset fault label map and identifies whether multi-factor composite anomalies exist. This enables the invention to handle scenarios where multiple faults occur simultaneously or where latent factors overlap, significantly improving the comprehensiveness of fault identification and the rationality of classification.

[0014] This invention achieves automatic linkage generation of response strategies and intelligent adjustment of sampling frequency, improving the system's adaptability and resource efficiency. After determining the fault type, the response strategy generation module retrieves strategy rules and writes operation instructions into the execution buffer. Combined with the identification results, the power channel acquisition cycle is dynamically adjusted. Fault judgment feedback is embedded in the control flow, realizing a closed-loop response between acquisition behavior and judgment results. This processing method breaks the limitations of traditional fixed acquisition cycle settings, enabling the system to flexibly adjust the sampling frequency according to the operating status of different channels, enhancing the ability to capture signals in abnormal states. Simultaneously, it reduces sampling density in stable channels to release resources, ultimately improving detection sensitivity while optimizing storage and communication overhead, demonstrating a highly adaptive data acquisition and control capability. Attached Figure Description

[0015] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of the adaptive fault assessment system for power acquisition terminals based on grey relational analysis in this invention. Detailed Implementation

[0016] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0017] Example The adaptive fault assessment system for power acquisition terminals based on grey relational analysis includes a data acquisition and processing module, a sequence standardization module, an anomaly factor identification module, a fault type determination module, and a response strategy generation module. The data acquisition and processing module collects operational data from multiple power channels, constructs corresponding time series for each, and processes the data. Simultaneously, it uses a stored normal operation standard sequence as a reference. This module serves as the initial step in the fault assessment process, undertaking the tasks of raw acquisition and structured organization of power information, ensuring that various parameters such as voltage, current, active power, and reactive power form complete and continuous operational records independently for each channel. By constructing time series, the dynamic change trend of each channel's data can be preserved. Furthermore, through time series alignment and data preprocessing, the acquired data meets the basic requirements of subsequent analysis modules for consistent input data length and standardized format. Using the normal operation standard sequence as a reference helps establish a stable benchmark for subsequent similarity calculations and anomaly deviation identification, thereby ensuring the overall identification results have reliable reference value and comparative consistency.

[0018] The sequence standardization module normalizes the values ​​of all processed running data sequences and the reference sequence, and calculates the grey relational coefficient at each corresponding sampling point, forming a correlation coefficient sequence to reflect the instantaneous similarity of each sampling channel at each time point. This module unifies the numerical scale of the data from each channel through normalization, eliminating the influence of dimensional differences and acquisition errors, so that subsequent correlation coefficient calculations are not affected by the magnitude of absolute values, focusing on trend changes. In the calculation of grey relational coefficients, the relative similarity between the running data and the reference sequence at each sampling point is compared to form a local similarity sequence on the time axis. The final correlation coefficient sequence provides basic data reflecting the subtle differences between the current behavior state and the normal state of the power channel, and has strong sensitivity to abnormal trends and signal disturbance identification capabilities.

[0019] The anomaly factor identification module averages the grey relational coefficient sequence over time to obtain the full-sequence correlation value for each electrical channel. When the correlation value of a certain item is lower than its corresponding set threshold, it extracts the low-correlation factor combination for the current period based on the corresponding channel information and electrical parameter type. All low-correlation factor combinations are then aggregated to form the correlation anomaly feature set for the current period. Based on the correlation coefficient sequence, this module extracts the similarity of the overall operating curve of each channel by performing an averaging operation over time, forming a full-sequence correlation value representing the behavior state of the entire period. The level of this value directly reflects the overall consistency between the current state of the channel and the normal reference state. By setting reasonable thresholds for different channels and parameter types, the correlation values ​​are classified and judged, effectively identifying abnormal channels exhibiting deviation trends within the current period. Combining channel indexes and electrical parameter classification information, structured low-correlation factor combinations are constructed, and these combinations form the correlation anomaly feature set for the current period, providing a high-dimensional, quantitative input feature foundation for subsequent fault type judgment.

[0020] The fault type determination module, based on a pre-built mapping relationship between abnormal features and fault labels, performs graph matching in the fault label space using the generated feature set to determine the specific fault type and whether multi-factor composite anomalies exist, prioritizing and classifying the types. This module establishes the association structure between features and labels during system initialization, mapping different combinations of abnormal factors to specific fault types through a graph approach. In actual operation, after obtaining the associated abnormal feature set for the current period, the most likely fault type is identified by matching the similarity between this set and the label index structure in the graph. This identification process has multi-label concurrent processing capabilities, enabling further determination of whether a multi-factor composite anomaly exists when multiple potential matching results exist. By setting response priorities among fault types, multiple possible faults are sorted by level to ensure that the system prioritizes responding to the dominant anomaly with the greatest impact on operation, while recording and classifying other secondary anomalies to improve overall fault understanding and learning capabilities.

[0021] The response strategy generation module, based on the determined fault type, retrieves the corresponding response flow from the strategy rule base, writes the operation instructions in the response flow into the execution buffer, and dynamically adjusts the acquisition cycle of the corresponding power channel. This module, based on the result from the fault type determination module, retrieves the strategy rule matching the type and obtains the response flow from the rule base, consisting of instructions such as acquisition scheduling, data re-acquisition, and frequency adjustment. Each strategy is assembled and written into the execution buffer, taking effect after system scheduling to ensure the timeliness and consistency of processing actions. Based on this, and combined with the correlation value of abnormal channels within the current cycle, its acquisition cycle is dynamically adjusted, thereby achieving enhanced sampling of important abnormal factors and appropriately reducing the sampling frequency of stable channels. This constitutes a fault-characteristic-driven dynamic acquisition rhythm control mechanism, enabling the terminal to have real-time adjustment and resource optimization capabilities.

[0022] Data processing refers to the time-series data being synchronized, missing data filled in, and numerically smoothed to form a sequence structure of equal length without abrupt changes. The time series refers to the sequence of operational data collected sequentially by the power acquisition and processing module for each power channel within a set sampling period. It reflects the dynamic changes of electrical parameters such as voltage, current, active power, and reactive power over time. Each time series corresponds to a combination of a channel and a type of electrical parameter, possessing both independence and continuity, and serves as the fundamental input for the subsequent sequence standardization module to calculate the grey relational coefficient.

[0023] Timing synchronization refers to the process where, due to sampling delays, data communication lags, or clock errors, the time axes of sampling points in different power channels may not be perfectly aligned. Timing synchronization calibrates the timestamps of all channels, ensuring that all time series within the same sampling period logically have the same time index. This guarantees effective comparison of data from different channels at the same time point.

[0024] Missing data imputation refers to the process in time series analysis where some sampling points may fail to acquire valid data due to momentary signal loss, communication interruption, or abnormal equipment response, creating data gaps. Through missing data imputation, the system automatically identifies the missing locations and fills them by interpolation estimation based on preceding and following valid points or by copying adjacent values, restoring each sequence to an uninterrupted and complete state.

[0025] Numerical smoothing refers to the process of smoothing time series data to prevent short-term abnormal fluctuations caused by external electromagnetic interference, sensor jitter, or sudden load changes during the original sampling process. To enhance the stability of the series, moving averages or low-pass filtering are used to smooth the series numerically. This process suppresses local spikes without changing the overall trend of the data, making the processed series more suitable for subsequent grey relational analysis.

[0026] Forming a sequence structure of equal length and without abrupt changes means that, after the above three steps, the time series corresponding to all power channels have the same length on the time axis, the same number of sampling points, and continuous numerical changes without abnormal breakpoints or abrupt changes, thus meeting the technical requirements of the sequence standardization module for the consistency and stability of the input structure. For example, in a certain power acquisition cycle, the voltage time series of channel 1 has two unresponsive sampling points, the active power series of channel 2 has a 1-second sampling delay at the beginning, and the reactive power series of channel 3 shows a short-term spike signal. After performing time synchronization, interpolation, and moving average processing on the three, the three sequences can be standardized into a data structure with consistent length, aligned sampling, and smooth waveform, thus entering the gray relational coefficient calculation process and providing stable input for the anomaly factor identification module.

[0027] The starting point zeroing method is used for numerical normalization, and the running data sequence after the power channel processing is defined as follows: ; Define the reference sequence as: m and n are both initial index numbers and have the same value; After processing using the initial point zeroing method, we obtain: ; ; in, It can be deduced that: ;at this time, The value can be 0 or 1. This is the index of the number after the initial point zeroing method.

[0028] Calculating the grey relational coefficient at each corresponding sampling point refers to: Within a fixed time window, assume the sequence Given an n-1 time series, then the sequence... In the interval The corresponding graph above should be a broken line. ,in , ; sequence Corresponding broken line In the interval The change in area can be expressed as: ; For sequence and In the interval The change in area on is denoted as and ; but ; For sequence and The grey relational coefficient, at this point, and If they are 0 and 1 respectively, then If the correlation coefficient is greater than zero, the correlation coefficient is positive, indicating a positive correlation between the sequences; if... If the correlation coefficient is less than zero, the correlation coefficient is negative, indicating a negative correlation between the sequences; if This indicates that the sequences are uncorrelated.

[0029] The anomaly identification module averages the grey relational coefficient sequence over time to obtain the full sequence correlation value for each power channel. After the time series of a power channel is normalized, the sequence standardization module calculates the grey relational coefficient for each sampling point of that channel. The grey relational coefficient describes the local similarity between the actual data of the channel at that sampling point and the corresponding data in the reference sequence. These sampling points form a grey relational coefficient sequence within a complete acquisition cycle, exhibiting a time series structure. Averaging over time means calculating the average of all grey relational coefficients corresponding to a power channel within the current cycle to obtain the overall similarity performance value of that channel in that cycle; this value is the full sequence correlation value for that power channel. The full sequence correlation value is a global indicator characterizing the degree of closeness between the overall operating trend of the channel and the reference state, used to measure the overall degree of abnormal deviation.

[0030] When the correlation coefficient of a certain parameter falls below its corresponding set threshold, it indicates that the operating behavior of that power channel in that cycle has systematically deviated from the reference state. The set threshold is a pre-configured judgment baseline, with different numerical ranges set for different channels or electrical parameters to accommodate the sensitivity characteristics of different power types. The threshold can be determined based on equipment characteristics, historical fluctuation ranges, or statistical patterns, and is used to control the false alarm rate and false negative rate during the identification process. Extracting the low correlation factor combination for the current cycle, combining its corresponding channel information and electrical parameter type, means that for all channels below the threshold, its channel number and corresponding electrical parameter type (such as voltage, active power, reactive power, etc.) are extracted to form a pair of structured data, indicating that the channel's behavior deviation in the current parameter item has reached the warning level. This structure is the low correlation factor combination, the smallest unit for locating the source of the anomaly. Summarizing all low correlation factor combinations means uniformly collecting all channel parameter items with correlation coefficients below the threshold in the current cycle, organizing them according to a standard structure, removing duplicates, and generating a correlation anomaly feature set. The associated anomaly feature set, as a structured set, contains all channel-parameter pairs that deviate from the reference sequence within the current period, and is used to perform spectrum matching operations in the fault type determination module.

[0031] For example, within a certain period, the correlation between the voltage sequence of channel 3 and the reference sequence is 0.62, which is lower than the voltage-class threshold of 0.7; the correlation between the active power of channel 5 and the active power of channel 5 is 0.55, which is lower than the corresponding threshold of 0.65; the correlation between the active power of channel 3 and the active power of channel 5 within this period is 0.81, which does not trigger the threshold. Therefore, this period forms two low-correlation factor combinations, namely "channel 3-voltage" and "channel 5-active power", which constitute the correlation anomaly feature set for this period, providing input for subsequent fault label identification.

[0032] The fault type determination module, based on a pre-built mapping relationship between abnormal features and fault labels, performs graph matching in the fault label space using the generated feature set to determine the specific fault type and whether there are multi-factor composite anomalies, prioritizing and classifying the types. During the fault type determination process, the associated abnormal feature set identifies fault type labels through graph matching. The graph matching process includes calculating a deviation score for each preset label and filtering out all labels with scores below a preset recognition threshold based on the score results, forming the identifiable label set for the current period.

[0033] Before the above-mentioned graph matching is performed, the mapping relationship between abnormal features and fault labels needs to be constructed during the system initialization phase. Abnormal features refer to a structured set of all low-correlation factor combinations extracted by the abnormal factor identification module within the current period. Each low-correlation factor combination consists of an electrical channel number and an electrical parameter type, serving as the basic element characterizing a certain type of fault. Fault labels refer to the coded identifier corresponding to each type of fault in the system-defined set of identifiable fault types, such as communication faults, voltage deviations, power anomalies, and illegal wiring. Each fault label is associated with a set of typical low-correlation factor combinations.

[0034] The mapping relationship between abnormal features and fault labels is constructed as follows: Based on the collection periods in historical operational data records that have been manually or intelligently identified and labeled as a specific fault type, the combinations of low-correlation factors that actually occur within those periods are statistically analyzed. Commonly occurring channel-parameter combinations are extracted from multiple sample periods to establish a unique abnormal feature index structure for each fault label. This structure is the label index sequence, defining a template for representative and repetitive abnormal factor combinations for each type of fault in historical data. To enhance the generalization ability of the matching graph, each fault label can contain multiple abnormal feature sample combinations, with each combination forming a graph node. Within the graph, the same label can correspond to multiple abnormal behavior paths.

[0035] For example, typical characteristics of voltage offset fault labels include low correlation between "Channel 1-Voltage", "Channel 3-Voltage", and "Channel 5-Voltage" appearing simultaneously and frequently across multiple historical periods. Communication fault labels, on the other hand, exhibit data anomalies across multiple channels in multiple electrical parameter dimensions with irregular distribution. By encoding and storing this structured data using a channel-parameter index, a fault label space is formed. During graph matching, the current period's associated anomaly feature set is used as input, and a matching process is performed in the label space to determine which labels' corresponding features are most similar.

[0036] The identifiable tag set is used to represent all candidate fault types that meet the set requirements for matching abnormal features within the current period. If the set contains only one tag, then that tag is determined as the final fault type tag for the current period. If the set contains multiple tags, then no type decision is made for the time being, and the subsequent priority sorting and classification process begins.

[0037] The deviation scoring calculation mechanism comprises two consecutive calculation stages: the first stage is matching location identification, and the second stage is score calculation. In the first stage, for each low-correlation factor combination contained in the correlated anomaly feature set generated in the current period, its index position in the two-dimensional structure composed of the electrical quantity channel and electrical parameter type combination is extracted, forming the index sequence of features for this period, serving as the basis for spatial comparison. The electrical quantity channel refers to the physical channel number used for data acquisition, and the electrical parameter type refers to the type of electrical physical quantity represented by the acquired data, such as voltage, active power, reactive power, and frequency. The combination of these two forms a one-to-one corresponding two-dimensional logical index structure used to identify the spatial location information of the specific acquisition object. The index sequence refers to the set of numbers composed of channel and parameter pairs corresponding to all low-correlation factor combinations within the current period, representing the set of data locations where anomalies occurred within that period.

[0038] Subsequently, the index sequence is compared item by item with the label index sequence constructed during the initialization phase for each fault type label in the fault label space. The label index sequence is a set of typical low-association factor combinations statistically derived from historical training samples during the construction phase of the mapping relationship between abnormal features and fault labels. It represents the distribution pattern of abnormal factors that frequently occur in a specific fault type throughout historical periods. Item-by-item comparison refers to checking whether the index number of each element in the index sequence matches the corresponding element in the label index sequence in sequence. If they match, a match is considered to have occurred. The number of positions with completely identical indices is counted, and this number is used as the baseline for the matching degree score between the feature and the corresponding label in the current period. The matching degree score baseline is used to represent the degree of similarity in spatial structure between the current abnormal feature and the specific fault type, and is an important prerequisite for performing the scoring calculation.

[0039] In the second stage, the current cycle correlation value corresponding to all factors with consistent index positions is extracted, and the deviation is calculated between this value and the average correlation value of the training samples corresponding to each label in the label index sequence. The current cycle correlation value is the average value obtained by averaging over time in the abnormal factor identification module, representing the overall deviation of the current power channel within the current cycle. The average correlation value of the training samples is the average correlation value of the channel parameter in the historical samples corresponding to the fault type, representing a reference value for the standard abnormal form. Each deviation value is weighted according to the factor importance weights set in the label space construction stage. The factor importance weights are used to reflect the influence of different channel-parameter combinations on the accuracy of fault type identification, and are usually set through historical misjudgment rate, sensitivity assessment, or expert experience. Finally, all weighted deviation values ​​are summed to form a complete deviation score, which is used to comprehensively quantify the difference between the current cycle characteristics and the fault type label.

[0040] The above calculation is repeated for all labels, filtering out all labels with deviation scores below the recognition threshold to form the set of identifiable labels for the current period. The recognition threshold is a boundary value set during system initialization, used to remove labels with low matching degree to ensure that the final retained candidate labels have sufficient feature similarity.

[0041] For example, in one cycle, the abnormal feature set contains three low-correlation factor combinations: "Channel 2 - Voltage," "Channel 3 - Active Power," and "Channel 5 - Voltage," corresponding to index sequences [2,1], [3,2], and [5,1]. During the matching process, the label index sequence for the communication fault label is [1,3], [3,2], [5,1], with two positions completely consistent; the label index sequence for the voltage offset label is [2,1], [5,1], [6,1], with two positions consistent. For each label, the difference between the current cycle correlation value and the training sample value at the two consistent positions is calculated, multiplied by their respective factor weights, and then summed to form a deviation score. Finally, if both label scores are below the recognition threshold, they are included in the set of identifiable labels.

[0042] When the set of identifiable labels contains multiple labels, a type priority sorting and classification operation is performed. The set of identifiable labels refers to the set of all fault labels with deviation scores below the identification threshold, representing all candidate fault types that highly match the associated abnormal feature set within the current period. The type priority sorting and classification operation refers to determining, based on preset priority rules, which label should be executed first in the fault handling response process when multiple candidate labels exist, to ensure that the response behavior is focused and non-conflicting.

[0043] During the sorting process, the priority parameter corresponding to each tag in the set of identifiable tags is read. The priority parameter is a response level value assigned to each fault tag during the tag space initialization phase, reflecting the urgency of responding to different types of faults in the system. This parameter is represented numerically, typically as a non-negative integer or floating-point number. A higher value indicates a greater impact on system security, stability, or business continuity once the fault type occurs; therefore, when multiple fault types occur concurrently, they should be responded to first. The parameter value is set based on a graded assessment of the impact of historical response delays on the system. This involves analyzing historical fault event data to statistically determine the severity of consequences such as anomaly expansion, equipment damage, or data loss caused by different faults under response delay conditions, thereby constructing a response importance model for the impact of fault types on the system.

[0044] The labels are sorted from highest to lowest parameter value, and the top-ranked label is determined as the dominant fault type for this period. The remaining labels are then classified as secondary response types in descending order of their ranking. The dominant fault type refers to the fault label with the highest priority parameter value in the multi-label identification results for the current period, indicating that the fault represented by this label should be the primary focus of processing in this period. Secondary response types refer to other labels ranked after the dominant type; their corresponding strategies are usually not implemented immediately, but they can still serve as supplementary reference records.

[0045] After sorting, only the response flow corresponding to the dominant fault type is used for control logic generation and written to the execution buffer. The execution buffer is a structure in the response strategy generation module used to temporarily store the execution instructions for the next cycle. In this structure, the response flow corresponding to the dominant fault type is loaded in the form of an instruction sequence, which is triggered for execution after the system enters the next response cycle. This approach avoids conflicts caused by multiple strategies triggering in the same cycle while retaining the identification results of secondary response types as reference records for subsequent data accumulation and supplementation of abnormal samples by the self-learning module. The self-learning module is an internal module that continuously collects, optimizes, and updates the graph matching mechanism and fault label space during long-term operation. By retaining multi-label identification information, it can improve the adaptability of the label identification model to complex or atypical fault scenarios.

[0046] For example, in a set of identifiable tags for a given period, tags for communication failure, voltage anomaly, and illegal wiring all match successfully, with corresponding priority parameters of 9, 7, and 4, respectively. After sorting, communication failure is identified as the dominant failure type, and its response strategy includes operations such as soft restarting the communication module and switching the uplink channel, which are written to the execution buffer. The other two tags are recorded in the fault event log. If the communication strategy fails to respond subsequently, the system can refer to the secondary response type to execute a backup strategy.

[0047] During the response strategy generation process, the acquisition instructions contained in the execution buffer undergo pre-scheduling processing. Based on the current cycle correlation degree value corresponding to each low correlation factor combination in the current cycle correlation anomaly feature set, the acquisition cycle of the mapped power channel is dynamically adjusted. The execution buffer is a data structure used to temporarily store strategy instructions to be executed. After fault identification and strategy generation are completed in the current cycle, all response actions are encapsulated as acquisition instructions or control instructions and written into this buffer, to be executed when the scheduling logic of the next cycle is triggered. Acquisition instructions refer to the set of commands used to control the power channel to perform data sampling, including information such as sampling object, sampling parameters, and sampling cycle. The current cycle correlation anomaly feature set is a structured set generated by the anomaly factor identification module, containing all channel and electrical parameter combinations that exhibit a gray correlation degree value lower than their corresponding set threshold in the current cycle, i.e., low correlation factor combinations. Each low correlation factor combination corresponds to a power channel and an electrical parameter type, and has a clear spatial location. The current cycle correlation degree value refers to the value obtained by averaging the gray correlation coefficient sequence in the time dimension, used to quantify the overall behavioral deviation of the channel in the current sampling cycle. Dynamically adjusting the sampling period of the mapped power channel refers to adjusting the sampling period parameter in the acquisition command based on the aforementioned correlation value, so as to realize the period adaptive control logic in the response strategy.

[0048] The adjustment method is as follows: For each acquisition command, check if its target channel has a matching item in the associated anomaly feature set. The target channel is the data channel number controlled by the acquisition command, and the matching item refers to whether the channel-parameter combination specified in the current acquisition command appears in the anomaly feature set. If it exists, it indicates that the object controlled by the command is currently exhibiting abnormal behavior.

[0049] If it exists, the current cycle correlation value corresponding to the channel's electrical parameter combination is extracted, and this value is compared with the preset cycle adjustment judgment interval. The cycle adjustment judgment interval is a numerical interval structure set during the system design phase, typically containing three segments: a lower limit interval, a hold interval, and an upper limit interval, used to control the adjustment decisions of shortening, keeping unchanged, or extending the cycle, respectively. If the value is below the lower limit interval, the channel's acquisition cycle is shortened to a certain proportion of the original cycle to enhance the time capture frequency of abnormal fluctuation signals. The original cycle refers to the reference sampling cycle set in the acquisition command during initialization or the previous cycle. Shortening the cycle is usually calculated using a fixed proportion (e.g., 0.5 times) to double the sampling frequency and improve the sensitivity to detecting severe fluctuations. If the value is below the upper limit interval but does not exceed the lower limit, the original acquisition cycle is kept unchanged to maintain normal acquisition. This situation indicates that although the channel exhibits some fluctuations, it has not exceeded the warning threshold and does not trigger adjustment. If the value is above the normal range, the acquisition cycle is multiplied by a preset extension coefficient to obtain the adjusted acquisition cycle, used to release data channel load resources and reduce storage overhead. An extension factor is a floating-point value greater than 1, such as 1.5 or 2, which indicates that the sampling period is extended proportionally to reduce the sampling frequency on stable channels and improve the overall resource utilization efficiency of the system.

[0050] All cycle adjustment decisions are completed before the execution buffer scheduling and embedded in the response process as part of the power acquisition and control behavior for this cycle, forming a closed-loop response structure. The closed-loop response structure means that fault identification, response strategy generation, and acquisition behavior adjustment achieve data feedback and control closure within the same cycle, improving the real-time performance of fault handling and the system's adaptability.

[0051] For example, if "Channel 3 - Active Power" appears in the abnormal feature set in a certain cycle, and its current cycle correlation value is 0.42, which is lower than the lower limit range of 0.50, then the sampling cycle of the original 10-second cycle in the acquisition command will be shortened to 5 seconds; another channel, "Channel 6 - Voltage", has a correlation of 0.68, which is within the holding range (0.55-0.75), and its sampling cycle remains unchanged; while "Channel 9 - Reactive Power" has a correlation of 0.92, which is higher than the upper limit range of 0.85, then the original 10-second sampling cycle command will be extended to 20 seconds. Finally, the three adjusted acquisition commands will be written into the execution buffer and synchronously scheduled for execution in the next cycle.

[0052] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0053] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0054] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0055] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0056] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. An adaptive fault assessment system for power acquisition terminals based on grey relational analysis, characterized in that, It includes a data acquisition and processing module, a sequence standardization module, an anomaly factor identification module, a fault type determination module, and a response strategy generation module; The data acquisition and processing module collects operational data from multiple power channels, constructs corresponding time series for each channel, and processes the data. At the same time, it calls the stored normal operation standard sequence as a reference object, i.e., the reference sequence. The sequence normalization module performs numerical normalization on all processed running data sequences and reference sequences, and calculates the grey relational coefficient at each corresponding sampling point to form a relational coefficient sequence, which is used to reflect the instantaneous similarity of each sampling channel at each time point. The abnormal factor identification module averages the grey relational coefficient sequence over time to obtain the full sequence correlation value for each power channel. When the correlation of a certain item is lower than its corresponding set threshold, it extracts the low correlation factor combination for the current period by combining its corresponding channel information and power parameter type. It then summarizes all low correlation factor combinations to form the correlation abnormal feature set for the current period. The fault type determination module is based on the pre-built mapping relationship between abnormal features and fault labels. It uses the generated feature set to perform graph matching in the fault label space to determine the specific fault type, and determines whether there are multi-factor composite anomalies, and sorts and classifies the types according to their priority. The response strategy generation module retrieves the corresponding response process from the policy rule base based on the identified fault type, writes the operation instructions in the response process into the execution buffer, and dynamically adjusts the collection cycle of the corresponding power channel.

2. The adaptive fault judgment system for power acquisition terminals based on grey relational analysis according to claim 1, characterized in that, Data processing refers to: Time series are synchronized, missing data are filled in, and numerical smoothing is performed to form a sequence structure of equal length and without abrupt changes.

3. The adaptive fault judgment system for power acquisition terminals based on grey relational analysis according to claim 2, characterized in that, The starting point zeroing method is used for numerical normalization, and the running data sequence after the power channel processing is defined as follows: ; Define the reference sequence as: m and n are both initial index numbers and have the same value; After processing using the initial point zeroing method, we obtain: ； ； in, It can be deduced that: ;at this time, The value can be 0 or 1. This is the index of the number after the initial point zeroing method.

4. The adaptive fault judgment system for power acquisition terminals based on grey relational analysis according to claim 3, characterized in that, Calculating the grey relational coefficient at each corresponding sampling point refers to: Within a fixed time window, assume the sequence Given an n-1 time series, then the sequence... In the interval The corresponding graph above should be a broken line. ,in , ; sequence Corresponding broken line In the interval The change in area can be expressed as: ； For sequence and In the interval The change in area on is denoted as and ; but ; For sequence and The grey relational coefficient, at this point, and They are 0 and 1 respectively.

5. The adaptive fault judgment system for power acquisition terminals based on grey relational analysis according to claim 4, characterized in that, During the fault type determination process, the associated abnormal feature set identifies fault type labels through graph matching. The graph matching process includes performing deviation score calculation for each preset label and filtering out all labels with scores lower than the preset identification threshold based on the score results, forming the identifiable label set for the current period. The identifiable tag set is used to represent all candidate fault types that meet the set requirements for matching abnormal features within the current period. If the set contains only one tag, then that tag is determined as the final fault type tag for the current period. If the set contains multiple tags, then no type decision is made for the time being, and the subsequent priority sorting and classification process begins.

6. The adaptive fault judgment system for power acquisition terminals based on grey relational analysis according to claim 5, characterized in that, The deviation scoring calculation mechanism consists of two consecutive calculation stages: the first stage is matching position identification, and the second stage is score calculation. In the first stage, for each low-correlation factor combination contained in the associated abnormal feature set generated in the current cycle, its index position in the two-dimensional structure composed of the power channel and the combination of power parameter types is extracted to form the index sequence of the feature of this cycle, which serves as the basis for spatial comparison. Then, the index sequence is compared item by item with the label index sequence constructed in the initialization stage for each fault type label in the fault label space. The number of positions with completely identical indices is counted, and this number is used as the baseline for the matching degree score between the feature of this cycle and the corresponding label. In the second stage, the correlation values ​​of the current period corresponding to all factors with consistent index positions are extracted, and the deviation is calculated with the average correlation value of the training samples corresponding to the corresponding position of each label in the label index sequence. Each deviation value is weighted according to the factor importance weight set in the label space construction stage. Finally, all weighted deviation values ​​are summed to form a complete deviation score. The above calculation is repeated for all labels, and all labels with deviation scores lower than the recognition threshold are selected to form the set of identifiable labels for the current period.

7. The adaptive fault judgment system for power acquisition terminals based on grey relational analysis according to claim 6, characterized in that, When the set of identifiable tags contains multiple tags, perform type priority sorting and classification operations. During the sorting process, read the priority parameter corresponding to each tag in the set of identifiable tags, arrange them from high to low according to the parameter value, determine the first tag in the sorting as the dominant fault type in this cycle, and classify the remaining tags into secondary response types in the sorting order. After sorting, only the response flow corresponding to the dominant fault type is used for control logic generation and written to the execution buffer.

8. The adaptive fault judgment system for power acquisition terminals based on grey relational analysis according to claim 7, characterized in that, During the response strategy generation process, the acquisition instructions contained in the execution buffer undergo pre-scheduling processing. Based on the current cycle correlation degree value corresponding to each low correlation factor combination in the current cycle correlation anomaly feature set, the acquisition cycle of the mapped power channel is dynamically adjusted. The adjustment method is as follows: For each acquisition command, check if there is a matching item in the associated abnormal feature set for its target channel. If there is, extract the current cycle correlation value corresponding to the electrical parameter combination of the channel and compare the value with the preset cycle adjustment judgment range. If the value is below the lower limit range, shorten the acquisition cycle of the channel to a certain proportion of the original cycle. If the value is below the upper limit but does not exceed the lower limit, the original collection period remains unchanged; If the value is above the normal range, the acquisition period will be multiplied by a preset extension coefficient to obtain the adjusted acquisition period.

Images