Current transformer error dynamic monitoring method and system

Abstract

The invention relates to the technical field of power system measurement, and discloses a current transformer error dynamic monitoring method and system.The current transformer error dynamic monitoring method comprises the steps that current transformer time sequence data and a system event log are obtained; constructing a time sequence causal graph to represent the time correlation between the event and the error change; identifying potential causal links by applying a counter causal model; designing a multi-world simulation engine to generate an anti-fact scene; quantifying a causal effect by comparing actual observation with an anti-fact simulation result; establishing a monitoring mechanism to track key trigger events in real time; generating a dynamic causal interpretation report and adjusting a compensation strategy; according to the method, the limitation of traditional correlation analysis is broken through, the causal relationship and the correlation can be accurately distinguished, the real triggering factor of the error change of the current transformer can be accurately identified, the false alarm rate and the missing report rate are reduced, and the accurate dynamic monitoring of the error of the current transformer is realized.

Description

Technical Field

[0001] This invention relates to the field of power system measurement technology, and more specifically, to a method and system for dynamic monitoring of current transformer errors. Background Technology

[0002] Current transformers, as crucial measuring devices in power systems, directly impact the safe operation and economic efficiency of the power system. During actual operation, the measurement errors of current transformers dynamically change with varying system conditions, including fluctuations in ratio and phase angle errors. These error variations may originate from the individual or combined effects of various system events, such as load fluctuations, system switching, short-circuit faults, and lightning strikes. Traditional current transformer error monitoring methods primarily rely on static verification and periodic testing, making it difficult to achieve real-time tracking and accurate identification of dynamic error changes.

[0003] Existing dynamic monitoring technologies primarily rely on correlation analysis, which identifies potential influencing factors by statistically analyzing the degree of correlation between system events and error changes. However, correlation analysis has significant technical limitations: when faced with multiple events in a complex situation, it struggles to distinguish between genuine causal relationships and mere correlations, easily misinterpreting coincidental events as causal connections; furthermore, this method cannot effectively handle interactions and temporal dependencies between events, leading to insufficient accuracy in identifying the root causes of error changes under complex operating conditions. These technical deficiencies result in frequent false alarms and missed alarms in existing monitoring systems, impacting the reliability of power system operation and maintenance.

[0004] Furthermore, existing technologies lack a deep understanding of the mechanisms underlying error changes and the ability to predict them. Traditional methods primarily employ a passive monitoring model, allowing only post-event analysis after error changes occur, and cannot provide proactive early warnings or preventative measures.

[0005] With the increasing complexity and intelligence of power systems, there is an urgent need for a current transformer error monitoring technology that can accurately identify the causal relationship of error changes, achieve precise dynamic monitoring, and have predictive and early warning capabilities, so as to meet the higher requirements of modern power systems for measurement accuracy and intelligent operation and maintenance. Summary of the Invention

[0006] This invention provides a method and system for dynamic monitoring of current transformer errors, which solves the technical problem that related technologies have difficulty distinguishing between true causal relationships and only correlations when faced with multiple events, resulting in the inability to accurately identify the true triggering factors of error changes.

[0007] This invention provides a method for dynamic monitoring of current transformer errors, comprising the following steps: Collect and preprocess current transformer timing measurement data and system event logs, including data cleaning, standardization and outlier handling; Feature extraction is performed on the preprocessed data to identify key time series features and event features; Construct a temporal cause-effect graph to represent the temporal relationship between events and error changes, and perform initial weight calculation; A time-series cause-effect graph is a directed graph structure in which nodes represent system events or current transformer error changes, directed edges represent possible causal relationships, and the weight of the edges reflects the strength of the causal relationship. The counter-causal model is applied to analyze the time-series causal graph to identify potential causal links and correlation strengths; The counter causal model includes: The event counter module establishes an event frequency statistics table, recording the number of times each system event occurs within the observation period, the time of its first occurrence, the time of its last occurrence, and the average interval between occurrences. The error change monitoring module establishes an error change record table to continuously record the occurrence time, magnitude, direction, and duration of error changes in the current transformer. The correlation analysis module establishes an event-error co-occurrence count matrix and uses a hierarchical time window technique to establish three time windows of different scales: short-term, medium-term, and long-term. The co-occurrence count matrix is ​​calculated for each time window, and the overall correlation strength is obtained by weighted summation. The probabilistic reasoning module calculates conditional probability and joint probability based on the counting matrix and event frequency statistics table, and uses Bayes' theorem to handle the joint impact probability of event sequences on error changes. The causal link screening module screens potential causal links based on statistical significance tests. It calculates the chi-square statistic for each pair of event-error relationships, and considers the causal link to be statistically significant when the chi-square value is greater than the critical value. By generating counterfactual scenarios through a multi-world simulation engine, we can simulate the occurrence and non-occurrence of specific events and assess their impact. The multi-world simulation engine includes a state representation module, an event intervention module, a state transition calculation unit, and an uncertainty quantification unit. It intervenes in specific events through event operation functions to create parallel state evolutions of actual and counterfactual paths. Counterfactual scenarios refer to hypothetical situations that differ from actual observations, constructed by artificially intervening in the occurrence of a specific event. The generation of counterfactual scenarios is based on the controlled variable method: keeping all other conditions constant, only changing the state of the target event, and observing the differences in the system response. By comparing the differences between actual observations and counterfactual simulation results, the causal effects of each event on the changes in current transformer error are quantified. Based on the strength of causal effects, key triggering events affecting the error of current transformers are monitored and identified in real time, and early warning information is generated.

[0008] Furthermore, the preprocessing includes: The amplitude of the primary and secondary current values ​​of the current transformer is normalized. The comparison value error and phase angle error are standardized. Encode the event types in the system event log, converting non-numerical event types into numerical categorical variables.

[0009] Furthermore, constructing a time-series causal graph with a directed weighted graph structure includes: The algorithm for identifying error change points decomposes the original error sequence using wavelet decomposition and Daubechies-4 wavelet basis functions. It then applies a soft threshold function to shrink the wavelet coefficients at each level, calculates the first difference of the denoised error sequence, and applies a double threshold to identify significant error change points. The system event logs are standardized using an event sequence standardization algorithm, which assigns a timestamp, event type code, event severity, and event duration attributes to each event, forming a structured event sequence. A graph node creation algorithm is used to create graph nodes for each error change point and system event, and a node index table is established. Each node has node ID, node type, timestamp, and node weight attributes. The maximum time lag parameter is set based on the physical response characteristics of the system using a time window constraint algorithm, and a directed edge creation algorithm is used to perform time sequence constraints and time window constraint checks to create directed edges that meet the constraint conditions. The weight of each directed edge is calculated using an edge weight calculation algorithm and a multi-factor weighted model, including conditional probability, time decay and statistical correlation terms. By employing edge pruning algorithms based on weight threshold pruning, statistical significance pruning, and redundant edge pruning, insignificant edges are removed to reduce graph complexity.

[0010] Furthermore, the multi-world simulation engine consists of the following four core components: Status representation module: Represents the power system at time... The state is represented as a multidimensional vector; Event intervention module: Enables intervention in specific events through event operation functions; State transition calculation unit: Used to predict the temporal evolution of the system state under the condition that a specific event occurs or does not occur. It calculates the system state at the next moment based on the state transition function; The state transition calculation unit includes: Physical constraint layer: Constructing a set of constraint conditions based on the fundamental laws of power systems; Historical data learning layer: Learns historical state transition patterns through time series analysis algorithms; Hybrid inference layer: integrates physical constraints and historical learning results; State verification layer: Checks whether the prediction results meet the system constraints. If the predicted state violates the constraints, the state value is adjusted by the least squares method to meet the constraints. Uncertainty Quantification Unit: Evaluates the reliability of simulation results, providing statistical reliability assurance for causal effect analysis; contains four calculation modules: Monte Carlo sampling module: assesses the impact of parameter uncertainty on results through repeated sampling; Bayesian estimation module: Calculates the posterior probability distribution of the prediction results based on Bayes' theorem; Sensitivity analysis module: Identifies the input parameters that have the greatest impact on the results using partial differential equations; Confidence Interval Generator: Provides confidence intervals for prediction results based on statistical distribution theory.

[0011] Furthermore, the state representation module performs the following preprocessing on the input heterogeneous data: Minimum normalization is applied to continuous parameters to restrict their range of values ​​to a specific interval; Sine and cosine transformations are applied to periodic parameters to avoid period jumps. Discrete parameters are processed using one-hot encoding.

[0012] Furthermore, the steps to quantify the causal effects of each event on the change in current transformer error include: Identify the points of error variation in the analysis target and the set of events that may be related to them; By using a multi-world simulation engine, we can obtain predicted error changes under real-world and counterfactual scenarios and create parallel state evolutions for real-world and counterfactual paths. The causal effect of an event on error change is calculated using the average treatment effect formula, which is the difference between the expected value of error change in the actual scenario and the expected value of error change in the counterfactual scenario. To perform robustness verification, the calculation was repeated multiple times under random initial conditions, and the confidence interval of the causal effect was calculated using statistical distribution theory. Z-Scre normalization is applied to the causal effects of all events to convert effect values ​​of different scales and units into comparable uniform dimensions.

[0013] Furthermore, the steps for real-time monitoring and identification of key triggering events affecting current transformer errors include: Establish a priority queue for causal event monitoring, giving priority to events with high causal effect intensity; The causal effect strength threshold for triggering alarms is dynamically adjusted based on the historical statistical characteristics of the system's operating status using an adaptive threshold algorithm. The sliding time window analysis algorithm determines the size of the time window and the sliding step size, calculates the local causal effect value for the data in each sliding window, and compares the causal effect values ​​of adjacent windows to detect the dynamic changes in causal relationships. The event combination effect evaluation algorithm includes a combination identification algorithm to detect frequently co-occurring event combinations, an interaction effect calculation algorithm to quantify the synergistic or antagonistic effects between events, and an analytic hierarchy process (AHP) algorithm to decompose event combinations into main effects and interaction effects.

[0014] Furthermore, the sliding time window analysis algorithm includes: Determine the time window size and sliding step size; Calculate the local causal effect value for the data within each sliding window; Compare the causal effect values ​​of adjacent windows to detect dynamic changes in causal relationships; When the change exceeds a preset threshold, a re-evaluation is triggered.

[0015] Furthermore, it also includes: Generate dynamic causal explanation reports, including causal link visualization charts of error changes; Provide error compensation suggestions for different types of events; Construct an error trend prediction model; Build a knowledge base, accumulate experience in causal analysis, and continuously optimize compensation strategies.

[0016] This invention discloses a dynamic monitoring system for current transformer errors, comprising: The data acquisition and preprocessing unit is used to acquire timing measurement data of the current transformer and system event logs, and to perform amplitude normalization, standardization and event type encoding conversion. The temporal cause-effect graph construction unit is used to construct a directed weighted graph structure. It represents the temporal causal relationship between system events and error changes through error change point identification algorithm, event sequence normalization algorithm, graph node creation algorithm, time window constraint algorithm, directed edge creation algorithm, edge weight calculation algorithm, and edge pruning algorithm. The counter causal model unit, based on event frequency statistics and time series correlation analysis, identifies a set of potential causal relationship candidates that have passed statistical significance tests through an event counter module, an error change monitoring module, a correlation analysis module, a probabilistic reasoning module, and a causal link screening module. The multi-world simulation engine, based on state space search and conditional reasoning, includes a state representation module, an event intervention module, a state transition calculation unit, and an uncertainty quantification unit. It intervenes in specific events through event operation functions to generate parallel state evolutions of actual and counterfactual paths. The causal effect quantification unit is used to calculate the causal effect of an event on error changes using the average treatment effect formula, perform robustness verification and calculate confidence intervals, and normalize the causal effect. The critical triggering event monitoring unit is used to establish a causal event monitoring priority queue. Through adaptive threshold algorithm and sliding time window analysis algorithm, it monitors and identifies critical triggering events that affect the error of current transformer in real time. The dynamic report generation unit is used to generate causal link visualization charts of error changes, provide error compensation suggestions for different types of events, and build error trend prediction algorithms and knowledge bases to accumulate causal analysis experience.

[0017] The beneficial effects of this invention are as follows: This invention overcomes the limitations of traditional correlation analysis and achieves accurate identification of causal relationships. Traditional current transformer error monitoring methods mainly rely on correlation analysis, which struggles to distinguish between genuine causal relationships and accidental correlations when faced with multiple events in a complex situation. This invention, by introducing counterfactual analysis theory and combining a multi-world simulation engine and a counter causal model, can effectively distinguish between events with only correlation and those with genuine causality, fundamentally solving the technical problem of inaccurate identification of the root causes of error changes.

[0018] This invention improves the accuracy and reliability of dynamic error monitoring. Through time-series causal graph construction and causal effect quantification algorithms, it can accurately identify key triggering events for changes in current transformer errors, significantly reducing false alarm and false negative rates. The multi-world simulation engine generates counterfactual scenarios, providing a reliable benchmark for causal relationship verification, thus significantly improving the accuracy of error tracking and providing a solid technical foundation for subsequent precise compensation.

[0019] This invention achieves intelligent monitoring capabilities that dynamically adapt to complex operating conditions. The adaptive threshold mechanism of this invention can dynamically adjust monitoring sensitivity according to the system's operating status; the sliding time window analysis algorithm can capture dynamic changes in causal relationships in real time; and the event combination effect evaluation model can handle the interaction of multiple events. These technological innovations enable the monitoring system to adapt to the complex and ever-changing operating conditions of the power system and maintain continuous and stable monitoring performance.

[0020] This invention enhances the intelligence and foresight of power system operation and maintenance. It not only monitors error changes in real time but also provides targeted compensation suggestions through dynamic causal explanation reports and predicts future error trends through error trend prediction models. The knowledge base's accumulation mechanism enables the system to continuously learn and optimize, achieving a shift from passive monitoring to proactive prevention, significantly improving the overall reliability of power system measurements and operation and maintenance efficiency. Attached Figure Description

[0021] Figure 1 This is a flowchart of a current transformer error dynamic monitoring method according to the present invention; Figure 2 This is a bar chart of the temporal causal graph edge weight analysis of the present invention; Figure 3 This is a line graph showing the comparative analysis of counterfactual scenarios of short-circuit faults according to the present invention; Figure 4 This is a line graph comparing the phase angle error in the counterfactual short-circuit fault scenario of the present invention; Figure 5 This is a bar chart showing the intensity analysis of causal effects of different system events in this invention; Figure 6 This is a grouped bar chart comparing the accuracy of error triggering factor identification in this invention; Figure 7 This is a grouped bar chart comparing the false alarm rate and the false negative rate of this invention. Detailed Implementation

[0022] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0023] At least one embodiment of the present invention discloses a method for dynamic monitoring of current transformer errors, such as... Figure 1 As shown, it includes the following steps: Step 1: Collect and preprocess current transformer timing measurement data and system event logs, including data cleaning, standardization and outlier handling; This step involves collecting time-series measurement data of the current transformer during its operation using power system monitoring devices. This data includes, but is not limited to, parameters such as primary current value, secondary current value, ratio error, and phase angle error. Simultaneously, system event logs are collected, including information on events that may affect the current transformer performance, such as load changes, system switching, and fault occurrences. The acquired data is synchronized according to timestamps to ensure the temporal consistency of the data analysis.

[0024] The raw data obtained needs to undergo the following preprocessing: The amplitude of the primary and secondary current values ​​is normalized to eliminate the influence of magnitude differences. The comparison error and phase angle error are standardized and converted into a standard distribution with a mean of 0 and a standard deviation of 1. Encode the event types in the system event log, converting non-numerical event types (such as "load change", "system switch" etc.) into numerical categorical variables; Timestamp data is converted into relative time intervals, which makes it easier to calculate the strength of temporal correlations.

[0025] like Figure 2 As shown, the correlation strength between different system events and error changes is illustrated. The edge weight values ​​allow for a direct comparison of which events are more strongly correlated with specific types of error changes. The figure shows that the edge weight for short-circuit faults and a sharp increase in phase angle error is the highest (0.882), indicating the strongest possible causal relationship between them.

[0026] Step 2: Extract features from the preprocessed data to identify key time series features and event features; Based on the acquired time-series data and event logs, a time-series cause-effect graph is constructed to represent the temporal relationship between events and error changes. The specific steps of the algorithm for constructing the time-series cause-effect graph are as follows: The error data of the current transformer is preprocessed, and significant error change points are identified after threshold detection and wavelet transform denoising. ,in, , , They represent the first , , One significant error change point; This represents the total number of significant error changes identified.

[0027] Threshold detection parameters include: Error change rate threshold: When the error change rate between two adjacent points exceeds the threshold, the point is considered to have a significant change. Error amplitude threshold: When the absolute value of the error exceeds the threshold, it is determined to be a point of significant change.

[0028] Wavelet transform parameters include: Wavelet basis function: db4 wavelet is selected; Number of decomposition layers: Usually 3-5 layers; Threshold function: A soft threshold function is used to shrink the coefficients.

[0029] Standardize the system event logs to form event sequences. ,in, , , They represent the first , , One event; Indicates the total number of events.

[0030] Create graph nodes for each error change point and system event; Set the maximum time lag parameter It is typically taken as 5 times the system response time; For each pair of events Sum of error change points ,if If the error change occurs within a valid time window after the event, then a directed edge is created. ; Calculate the initial weights of the edges using the following formula: ; in, Indicates from the event To error change Edge weights; Indicates an event With error change Frequency of co-occurrence (number of times / total observation time); Indicates an event The frequency of occurrence (number of events / total observation time); This is the time decay factor, which controls the rate of time decay; the larger the value, the faster the decay. Indicates error change The timestamp of the event; Indicates an event The timestamp of the event; This represents the time interval between the change in error and the occurrence of the event; This is a time decay function used to exponentially decay the association strength based on the time interval; the larger the time interval, the more significant the decay. Represents conditional probability, that is, in the event... Error changes under the conditions that occur The probability of occurrence.

[0031] Perform edge pruning, removing edges with weights below a threshold to reduce graph complexity.

[0032] Furthermore, the temporal causal graph construction algorithm supports a dynamic update mechanism, which can automatically adjust the graph structure and edge weights as new data is acquired.

[0033] Furthermore, the temporal causal graph construction algorithm can integrate domain knowledge constraints to exclude causal relationships that are physically impossible.

[0034] The completed time-series causal graph is a directed graph structure, where nodes represent system events or current transformer error changes, directed edges represent possible causal relationships, and the weights of the edges reflect the strength of the causal relationships.

[0035] like Figure 3 As shown, this diagram compares the changes in the current transformer ratio error over time in both real-world and counterfactual scenarios (assuming the fault did not occur) when a short-circuit fault occurs. This comparison provides a clear view of the actual impact of a short-circuit fault on the current transformer ratio error, validating the effectiveness of the counterfactual analysis method.

[0036] Step 3: Construct a temporal cause-effect graph to represent the temporal relationship between events and error changes, and perform initial weight calculation; Based on the constructed temporal causal graph, this step applies a counter causal model to identify potential causal links.

[0037] The counter-causal model is a causal inference algorithm based on event frequency statistics and time series correlation analysis. It establishes multiple counter variables to record co-occurrence patterns among events and combines probabilistic inference methods to identify causal relationships. Its core lies in transforming complex causal inference problems into computable statistical problems.

[0038] The counter causal model consists of the following five functional modules: Event counter module: Creates an event frequency statistics table. This table records each system event. Number of occurrences during the observation period First occurrence time Last occurrence time and average occurrence interval The specific calculation process is as follows: for each event type Iterate through the entire time series, and when this type of event is detected, the counter... Increase by 1, and record the timestamp at the same time.

[0039] Error Change Monitoring Module: Establishes an error change record table. This table continuously records the points of error change of the current transformer. Time of occurrence Variation range , change direction (Increase or decrease) and duration of change The identification of error change points is based on the moving average difference algorithm: calculating the difference sequence of error values ​​at continuous time points, and identifying points where the absolute value of the difference exceeds a preset threshold. When the time is reached, it is marked as a point of significant change.

[0040] Correlation Analysis Module: Establishing an Event-Error Co-occurrence Count Matrix ,in For the number of event types, This represents the number of error variation points. Matrix elements. Record events With error change within the specified time window The number of co-occurrences within a given timeframe. The calculation process is as follows: for each error change point... It searches for all events that have occurred within its forward time window and updates the corresponding count values. .

[0041] The correlation analysis module employs a hierarchical time window technique, establishing three time windows at different scales: short window (1-10 seconds): Capturing instantaneous causal relationships; Mid-term window (10-60 seconds): Capture the system response process; Long-term window (1-10 minutes): Capture the cumulative effect; For each time scale, calculate the short-term, medium-term, and long-term co-occurrence count matrices separately. , , Then, the overall correlation strength is obtained by weighted summation:

[0042] in , , , These represent the short-term, medium-term, and long-term counting weights, which are adjusted according to the specific application scenario. The default values ​​are 0.4, 0.3, and 0.3, respectively. For example, if instantaneous causal relationships are emphasized, these values ​​are increased. .

[0043] Probabilistic reasoning module: based on counting matrix Calculate conditional probability using event frequency statistics table. and joint probability The formula for calculating conditional probability is:

[0044] in For the event With error change The number of times they co-occur. For the event The total number of occurrences.

[0045] The probabilistic reasoning module handles temporal dependencies using Bayes' theorem: for event sequences Calculate its response to error change The probability of joint impact:

[0046] Causal Link Screening Module: Screens potential causal links based on statistical significance tests. For each pair of event-error relationships... Calculate its statistical significance index value:

[0047] in This represents the total number of observations. For the event Occurred but error changed Number of times it did not occur For error variation But the event occurred Number of times it did not occur For the event and error change The number of times that did not occur.

[0048] when Value greater than critical value (Usually, the significance level is taken) When the corresponding critical value is reached, the causal link is considered statistically significant and retained in the candidate set.

[0049] The output of the counter causal model is a set of potential causal relationship candidates that have undergone statistical significance testing. ,in For correlation strength, For statistical significance; Furthermore, the probabilistic reasoning module of the counter causal model can handle temporal dependencies and consider the potential interactions between events, thereby improving the accuracy of causal link identification.

[0050] The output of the applied counter causal model is a candidate set of potential causal relationships, which includes possible causal associations between system events and error changes.

[0051] like Figure 4 As shown, the graph compares the changes in the phase angle error of the current transformer over time in the actual scenario and the counterfactual scenario (assuming no fault occurs) when a short-circuit fault occurs. The graph clearly shows that a short-circuit fault causes a significant peak in the phase angle error, while the phase angle error remains stable and low in the counterfactual scenario. This further proves that the short-circuit fault is the main cause of the change in phase angle error.

[0052] Step 4: Apply the counter causal model to analyze the time-series causal graph to identify potential causal links and correlation strengths; This step designs and implements a multi-world simulation engine to generate counterfactual scenarios. The technical meaning and specific implementation of the multi-world simulation engine: A multi-world simulation engine is a computational system based on state-space search and conditional reasoning. It simulates system behavior under different event intervention conditions by constructing multiple parallel system state evolution paths. Its core technical idea is to expand a single observed timeline into multiple hypothetical timelines, each corresponding to a specific event intervention scenario.

[0053] The technical meaning of counterfactual scenarios: Counterfactual scenarios refer to hypothetical situations that differ from actual observations, constructed by artificially intervening in the occurrence of a specific event (presuming that an event that has already occurred did not occur, or presuming that an event that did not occur did occur). The generation of counterfactual scenarios is based on the controlled variable method: keeping all other conditions constant, only changing the state of the target event, and observing the differences in the system's response.

[0054] The multi-world simulation engine consists of the following four core components: 1. Status Representation Module: Represents the power system at any given time. The state is represented as a multidimensional vector ,in The state dimension is defined as follows. The state vector includes current transformer parameters (primary current, secondary current, ratio error, phase angle error), system load status (load size, load type, load change rate), environmental factors (temperature, humidity, electromagnetic interference intensity), and operating parameters (voltage level, frequency, phase).

[0055] This module performs normalization preprocessing on the input heterogeneous data: For continuous parameters Perform min-max normalization: This restricts the range of values ​​to the interval [0,1]. For periodic parameters Perform sine-cosine transformation: ,in The period length is used to avoid abrupt changes at the period boundaries; For discrete parameters Perform one-hot encoding: Discrete-value conversion A binary vector of 1s, where only one element is 1 and the rest are 0s.

[0056] 2. Event Intervention Module: Utilizes event handling functions. This allows for intervention in specific events. The event intervention module maintains an event intervention table. This table records each event. Intervention status ,in: This indicates that the event is forcibly removed (even if it actually occurred, it is set to not have occurred). This indicates that the actual state of the event is being maintained; This indicates that the event is forcibly added (even if it has not actually occurred, it will be set as occurred); The specific implementation process of event intervention: given the original event sequence and intervention operations Generate a sequence of events following the intervention. .when When removing an event, all events of type 1 are removed from the original sequence. The event; when When adding an event, insert the event at the specified time. .

[0057] 3. State Transition Calculation Unit: Used to predict the temporal evolution of the system state under the condition that a specific event occurs or does not occur. This unit is based on the state transition function. Calculate the system state at the next moment:

[0058] in for The vector of intervention events at any given time. This is the system parameter vector.

[0059] The state transition computation unit contains four processing levels: Physical constraint layer: Constructing a set of constraints based on fundamental laws of power systems (Ohm's law, Kirchhoff's laws, etc.) This ensures that the state transition meets physical feasibility requirements. Constraints include power balance constraints, voltage limit constraints, and current limit constraints.

[0060] Historical Data Learning Layer: This layer learns historical state transition patterns through time series analysis algorithms. It uses a sliding time window method based on historical state sequences. predict The changing trend.

[0061] Hybrid Inference Layer: This layer integrates physical constraints and historical learning outcomes. The integration formula is as follows: ,in For weight fusion.

[0062] State verification layer: Verifies whether the prediction results meet the system constraints. If the predicted state violates the constraints, the state value is adjusted using the least squares method to make it meet the constraints.

[0063] 4. Uncertainty Quantification Unit: This unit assesses the reliability of simulation results, providing statistical reliability assurance for causal effect analysis. It comprises four computational modules: Monte Carlo sampling module: Evaluates the impact of parameter uncertainty on results through repeated sampling. For each uncertain parameter... Assuming it follows a known distribution ,generate random samples The state transition calculation is performed for each sample to obtain the result distribution.

[0064] Bayesian estimation module: Calculates the posterior probability distribution of the prediction results based on Bayes' theorem. Given observation data... and prior distribution Calculate the posterior distribution of the parameters: .

[0065] Sensitivity analysis module: Identifies the input parameters that have the greatest impact on the results using partial differential equations. For the output... and input parameters Calculate sensitivity index , The larger the value, the more important the parameter. The more significant the impact on the results.

[0066] Confidence Interval Generator: Provides confidence intervals for prediction results based on statistical distribution theory. For predicted values... Calculate its standard error Then construct a confidence level of Confidence interval: ,in for The critical value of the distribution.

[0067] The operation of the multi-world simulation engine: for each point of interest in error variation The engine creates two parallel state evolution paths: the actual path (preserving the original state of all events) and the counterfactual path (for specific events). Perform intervention operations or Both paths start from the same initial state. Initially, the events are evolved separately at the same time step, resulting in different error change predictions, thus quantifying the events. For error changes The causal effect.

[0068] The multi-world simulation engine uses parallel computing to generate multiple possible counterfactual scenarios for each point of interest in error change, simulating whether the error change would still occur if a certain event had not occurred.

[0069] like Figure 5 As shown, the chart illustrates the causal effect strength of different system events on the change in current transformer error, and ranks the events based on normalized causal effect strength values. The chart clearly shows that short-circuit faults, system switching, and lightning interference have the highest causal effect strength, which provides an important basis for event priority ranking in current transformer error monitoring.

[0070] Step 5: Generate counterfactual scenarios using a multi-world simulation engine to simulate the occurrence and non-occurrence of specific events and assess their impact.

[0071] This step quantifies the causal effect of each event on the change in current transformer error by comparing the differences between actual observations and counterfactual simulation results. The execution steps of the causal effect quantification algorithm are as follows: Determine the points of change in the target error in the analysis. and its potentially related set of events ,in, , , They represent the first , , One event; Indicates the total number of events.

[0072] For each event The error change prediction values ​​are obtained from the multi-world simulation engine under the actual scenario (event occurs) and the counterfactual scenario (event does not occur); Calculate the average treatment effect (ATE): ; in, Indicates an event For error changes The average treatment effect; This represents the i-th system event; This represents the j-th error change point; This indicates that intervention can make the event... The operation that occurred; This indicates that intervention can make the event... Operations that do not occur; This is a conditional expectation function used to calculate the error change under specific intervention conditions. The average or expected value; It indicates the direction of a causal relationship, pointing from cause to effect.

[0073] To perform robustness verification, steps 2-3 were repeated multiple times under random initial conditions to calculate the confidence interval of the causal effect. The causal effects of all events are normalized, and the effect values ​​of different scales and units are converted into comparable uniform units through Z-Score standardization, and then an attribution distribution map is constructed. Based on preset thresholds, key events with significant causal effects are selected.

[0074] Furthermore, causal effect quantification algorithms can handle the complex causal effects of multiple events recursively, taking into account the sequential dependencies between events.

[0075] Furthermore, the causal effect quantification algorithm uses time decay weights when dealing with long-term causal effects, giving recent events higher weight values.

[0076] By quantifying causal effects, this method can distinguish between events that are merely correlated and those that are truly causal, thereby identifying the true triggering factors for changes in current transformer errors.

[0077] like Figure 6 As shown, the identification accuracy of traditional correlation analysis methods and the counterfactual analysis method proposed in this application is compared under different combinations of error types and event types. The charts visually demonstrate that the accuracy of this method is significantly higher than that of traditional methods in all test scenarios, especially in the case of slow-moving influencing factors that are difficult to identify, such as changes in ambient temperature.

[0078] Step 6: Compare the differences between actual observations and counterfactual simulation results to quantify the causal effects of each event on the change in current transformer error; Based on the quantified strength of causal effects, this step establishes a continuously operating monitoring mechanism to identify and track key triggering events affecting current transformer errors in real time. This mechanism includes: Establish a priority queue for causal event monitoring, giving priority to events with high causal effect intensity; Implement an adaptive threshold mechanism to dynamically adjust the causal effect strength threshold for triggering alarms based on the system's operating status; The specific steps for implementing the sliding time window analysis algorithm are as follows: Determine the size of the time window and the sliding step, which are typically 1-3 times the system response time; Calculate the local causal effect value for the data within each sliding window; Compare the causal effect values ​​of adjacent windows to detect dynamic changes in causal relationships; When the change exceeds a preset threshold, a re-evaluation is triggered; Integrated event combination effect assessment model, which includes: Combination recognition module: Detects frequently co-occurring combinations of events; Interaction effect calculation unit: quantifies the synergistic or antagonistic effects between events, and standardizes and performs feature cross-processing on the interaction characteristics of different event types; Hierarchical Analysis module: Decomposes event combinations into primary effects and interaction effects; Visual interface: Transforms complex interactive relationships into intuitive visual representations.

[0079] The key trigger event monitoring mechanism enables real-time and accurate tracking of current transformer error changes, providing a reliable basis for subsequent error compensation.

[0080] like Figure 7 As shown, the performance of the traditional correlation analysis method and this counterfactual analysis method is compared in four key indicators: false alarm rate, missed alarm rate, alarm accuracy rate, and overall monitoring effectiveness. The charts clearly demonstrate the significant advantages of this method in reducing false alarm rate and missed alarm rate, proving that counterfactual analysis can more accurately identify the true triggering factors of current transformer error changes, reduce unnecessary maintenance costs, and not overlook important issues.

[0081] Step 7: Based on the strength of causal effects, monitor and identify key triggering events affecting the error of the current transformer in real time, and generate early warning information; This step, based on the aforementioned analysis, generates a dynamic causal explanation report and adjusts the compensation strategy for the root cause in real time. Specifically, it includes: Automatically generate causal chain visualization charts of error changes to intuitively show the causal relationship between events and errors; Provides error compensation suggestions for different types of events, including parameter adjustment strategies and algorithm correction methods; An error trend prediction model is constructed, which consists of the following components: Temporal feature extractor: captures the temporal patterns and periodic features of historical error changes, and performs seasonal adjustment and trend decomposition preprocessing on the input error time series data; Causal Enhancement Prediction Unit: Integrates identified causal relationships to improve prediction accuracy; Multi-step predictor: capable of predicting error trends at different time scales; Confidence assessment module: Provides a reliability estimate for each prediction to facilitate decision-making; Build a knowledge base, accumulate experience in causal analysis, and continuously optimize compensation strategies.

[0082] With the dynamic cause-and-effect explanation report, system maintenance personnel can clearly understand the root cause of the error change in the current transformer and take more precise compensation measures.

[0083] Application example of this implementation method: Application scenarios: This embodiment is based on a dynamic monitoring system for current transformer errors in a 500kV substation. This substation is equipped with multiple current transformers to measure the current in equipment such as the main transformer, transmission lines, and busbars. In actual operation, the substation frequently faces complex events such as load fluctuations, lightning strikes, short-circuit faults, and system switching. These events cause dynamic changes in the current transformer measurement errors. Traditional monitoring methods struggle to accurately identify which events are the true triggers for error changes under these complex conditions.

[0084] In this embodiment, a current transformer (CT-L231) installed on a 500kV transmission line was selected for monitoring. The observation period was 30 days, during which various system events and current transformer error changes were recorded.

[0085] Implementation process example: Data acquisition and preprocessing:

[0086] This embodiment uses an automated measurement system in a substation to collect the primary and secondary current values ​​of the current transformer CT-L231, and calculates the ratio error and phase angle error. Simultaneously, system event logs are collected from the substation event management system. Table 1 shows a partial example of the collected raw data: Table 1: Example of current transformer timing measurement data and system event log;

[0087] The collected data is preprocessed as follows: the amplitude of the primary and secondary current values ​​is normalized; the comparison error and phase angle error are standardized; and the system event types are encoded and converted into numerical categorical variables.

[0088] Construction of time-series cause-effect graphs and application of counter-based cause-effect models: Based on the preprocessed data, a time-series cause-effect graph is constructed to represent the temporal correlation between system events and error changes. By identifying points of significant error change and analyzing event sequences, directed edges are established and their weights are calculated. In practical applications, a maximum time lag parameter is set. The time is 30 seconds (5 times the system response time). Table 2 shows an example of edge weight calculation for a time-series cause-effect graph: Table 2: Example of edge weight calculation for a temporal cause-effect graph;

[0089] After calculating the edge weights, perform an edge pruning operation to remove edges with weights below the threshold (0.5) to reduce the complexity of the graph.

[0090] A counter-based causal model was applied to analyze the time-series causal graph to identify potential causal links. This process included: an event counter module recording and tracking the frequency and timestamps of system events; an error change monitoring module continuously recording error change points and magnitudes; a correlation analysis module calculating the spatiotemporal correlation between events and error changes; a probabilistic inference module calculating conditional probabilities based on Bayesian networks; and a causal link screening module filtering potential causal links based on statistical significance thresholds. The analysis revealed the following potential causal relationships with high statistical significance: Short circuit fault → rapid increase in ratio error → rapid increase in phase angle error; System switching → Phase angle error increases; Increased load leads to a slow increase in ratio error.

[0091] Multi-world simulation engine generates counterfactual scenarios: For the selected potential causal relationships, counterfactual scenarios are generated through a multi-world simulation engine. This engine consists of four key components working together: a state representation module that represents the power system state as a multi-dimensional vector (including current transformer parameters, load states, etc.); an event intervention module that creates hypothetical scenarios to "remove" or "add" specific events; a state transition model (containing a physical constraint layer, a historical data learning layer, a hybrid inference layer, and a state verification layer) that predicts the evolution of the system state; and an uncertainty quantification unit (containing a Monte Carlo sampling module, a Bayesian estimation module, a sensitivity analysis module, and a confidence interval generator) that evaluates the reliability of the simulation results.

[0092] For example, considering the causal relationship of "short-circuit fault → sharp increase in ratio error," the system state evolution is generated under two scenarios: short-circuit fault occurrence and non-occurrence. Table 3 shows the counterfactual simulation results for a single short-circuit fault scenario: Table 3: Comparison of counterfactual simulation results for short-circuit faults;

[0093] As can be seen from Table 3, in the actual scenario where a short circuit fault occurs, both the ratio error and the phase angle error increase significantly; while in the counterfactual scenario (assuming that a short circuit fault does not occur), the error changes are very small, which is consistent with the slow change trend under normal operating conditions.

[0094] Quantifying causal effects and monitoring key triggering events: By comparing the differences between actual observations and counterfactual simulation results, the causal effects of each event on the change in current transformer error are quantified. The average treatment effect (ATE) is calculated, and the causal effects of all events are normalized to identify key events with significant causal effects. Table 4 shows the quantification results of the causal effect strength of different events: Table 4: Quantification results of the causal effect intensity of different system events;

[0095] Based on the quantified causal effect strength, a monitoring mechanism is established to identify and track key triggering events affecting current transformer errors in real time. This monitoring mechanism includes: establishing a priority queue for causal event monitoring, prioritizing events with high causal effect strength; implementing an adaptive threshold mechanism to dynamically adjust the alarm trigger threshold based on system operating status (currently set to 0.6 for causal effect strength); implementing a sliding time window analysis algorithm, setting the window size to 15 seconds (3 times the system response time), calculating local causal effect values ​​and comparing adjacent windows; and integrating an event combination effect evaluation model, quantifying the synergistic or antagonistic effects between events through a combination identification module, interaction effect calculation unit, hierarchical analysis module, and visualization interface.

[0096] The system automatically establishes a priority queue for causal event monitoring, prioritizing short-circuit faults, system switching, and lightning strikes—three types of events with high causal intensity. When these events occur, the system immediately increases the monitoring frequency of current transformer errors and prepares corresponding compensation strategies.

[0097] Technical effectiveness verification: This embodiment verifies the two main technical effects of the method through 30 days of substation current transformer operation monitoring: improved accuracy in identifying error triggering factors and reduced false alarm rate.

[0098] The accuracy of error trigger factor identification has been significantly improved: Table 5 shows a comparison of the accuracy of this method and the traditional correlation analysis method in identifying current transformer error triggering factors: Table 5: Comparison of accuracy rates for identifying error triggering factors;

[0099] Data shows that this method achieves an average accuracy of 90.0% in identifying the true triggering factors for changes in current transformer errors, an improvement of 29.5 percentage points compared to traditional correlation analysis methods. The accuracy improvement is particularly significant, reaching 37.4 percentage points, in identifying slowly influencing factors such as changes in ambient temperature.

[0100] Both false alarm rate and false negative rate decreased: During 30 days of operation monitoring, the system recorded 278 current transformer error change events, of which 126 were significant change events. Table 6 shows a comparison between this method and the traditional method in terms of false alarm rate and false negative rate: Table 6: Comparison of false alarm rate and false negative rate;

[0101] Data shows that after adopting this method, the false alarm rate decreased from 23.8% to 6.5%, and the false negative rate decreased from 18.7% to 4.2%, improving the overall monitoring efficiency of the system by 22.7 percentage points. This means that the system can more accurately identify error change events that truly require attention, avoiding unnecessary maintenance costs while ensuring that no important issues are overlooked.

[0102] Generate dynamic causal explanation reports: In this embodiment, the system also generates a dynamic causal explanation report based on the aforementioned analysis and adjusts the compensation strategy for the root cause in real time.

[0103] The system also provides error compensation suggestions for identified key events, such as: for errors caused by short-circuit faults, it is recommended to recalibrate parameters after the fault is restored; for phase angle errors caused by system switching, it is recommended to use an adaptive angle compensation algorithm.

[0104] The system also constructs an error trend prediction model, which consists of a time-series feature extractor (which first performs seasonal adjustment and trend decomposition on the input error time-series data), a causal enhancement prediction unit, a multi-step predictor, and a confidence assessment module. Using this model, the system can predict the error change trends that various events may cause within the next 24 hours and provide a 95% confidence interval.

[0105] The accumulated causal analysis experience of the system is stored in the knowledge base, and the compensation strategy is continuously optimized, which improves the accuracy and reliability of power system measurements.

[0106] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for dynamic monitoring of current transformer error, characterized in that, Includes the following steps: Collect and preprocess current transformer timing measurement data and system event logs, including data cleaning, standardization and outlier handling; Feature extraction is performed on the preprocessed data to identify key time series features and event features; Construct a temporal cause-effect graph to represent the temporal relationship between events and error changes, and perform initial weight calculation; A time-series cause-effect graph is a directed graph structure in which nodes represent system events or current transformer error changes, directed edges represent possible causal relationships, and the weight of the edges reflects the strength of the causal relationship. The counter-causal model is applied to analyze the time-series causal graph to identify potential causal links and correlation strengths; The counter causal model includes: The event counter module establishes an event frequency statistics table, recording the number of times each system event occurs within the observation period, the time of its first occurrence, the time of its last occurrence, and the average interval between occurrences. The error change monitoring module establishes an error change record table to continuously record the occurrence time, magnitude, direction, and duration of error changes in the current transformer. The correlation analysis module establishes an event-error co-occurrence count matrix and uses a hierarchical time window technique to establish three time windows of different scales: short-term, medium-term, and long-term. The co-occurrence count matrix is ​​calculated for each time window, and the overall correlation strength is obtained by weighted summation. The probabilistic reasoning module calculates conditional probability and joint probability based on the counting matrix and event frequency statistics table, and uses Bayes' theorem to handle the joint impact probability of event sequences on error changes. The causal link screening module screens potential causal links based on statistical significance tests. It calculates the chi-square statistic for each pair of event-error relationships, and considers the causal link to be statistically significant when the chi-square value is greater than the critical value. By generating counterfactual scenarios through a multi-world simulation engine, we can simulate the occurrence and non-occurrence of specific events and assess their impact. The multi-world simulation engine includes a state representation module, an event intervention module, a state transition calculation unit, and an uncertainty quantification unit. It intervenes in specific events through event operation functions to create parallel state evolutions of actual and counterfactual paths. Counterfactual scenarios refer to hypothetical situations that differ from actual observations, constructed by artificially intervening in the occurrence of a specific event. The generation of counterfactual scenarios is based on the controlled variable method: keeping all other conditions constant, only changing the state of the target event, and observing the differences in the system response. By comparing the differences between actual observations and counterfactual simulation results, the causal effects of each event on the changes in current transformer error are quantified. Based on the strength of causal effects, key triggering events affecting the error of current transformers are monitored and identified in real time, and early warning information is generated.

2. The method for dynamic monitoring of current transformer error according to claim 1, characterized in that, Preprocessing includes: The amplitude of the primary and secondary current values ​​of the current transformer is normalized. The comparison value error and phase angle error are standardized. Encode the event types in the system event log, converting non-numerical event types into numerical categorical variables.

3. The method for dynamic monitoring of current transformer error according to claim 1, characterized in that, Constructing a time-series causal graph with a directed weighted graph structure includes: The algorithm for identifying error change points decomposes the original error sequence using wavelet decomposition and Daubechies-4 wavelet basis functions. It then applies a soft threshold function to shrink the wavelet coefficients at each level, calculates the first difference of the denoised error sequence, and applies a double threshold to identify significant error change points. The system event logs are standardized using an event sequence standardization algorithm, which assigns a timestamp, event type code, event severity, and event duration attributes to each event, forming a structured event sequence. A graph node creation algorithm is used to create graph nodes for each error change point and system event, and a node index table is established. Each node has node ID, node type, timestamp, and node weight attributes. The maximum time lag parameter is set based on the physical response characteristics of the system using a time window constraint algorithm, and a directed edge creation algorithm is used to perform time sequence constraints and time window constraint checks to create directed edges that meet the constraint conditions. The weight of each directed edge is calculated using an edge weight calculation algorithm and a multi-factor weighted model, including conditional probability, time decay and statistical correlation terms. By employing edge pruning algorithms based on weight threshold pruning, statistical significance pruning, and redundant edge pruning, insignificant edges are removed to reduce graph complexity.

4. The method for dynamic monitoring of current transformer error according to claim 1, characterized in that, The multi-world simulation engine consists of the following four core components: Status representation module: Represents the power system at time... The state is represented as a multidimensional vector; Event intervention module: Enables intervention in specific events through event operation functions; State transition calculation unit: used to predict the temporal evolution of the system state under the condition that a specific event occurs or does not occur; calculates the system state at the next moment based on the state transition function; The state transition calculation unit includes: Physical constraint layer: Constructing a set of constraint conditions based on the fundamental laws of power systems; Historical data learning layer: Learns historical state transition patterns through time series analysis algorithms; Hybrid inference layer: integrates physical constraints and historical learning results; State verification layer: Checks whether the prediction results meet the system constraints. If the predicted state violates the constraints, the state value is adjusted by the least squares method to meet the constraints. Uncertainty Quantification Unit: Evaluates the reliability of simulation results, providing statistical reliability assurance for causal effect analysis; contains four calculation modules: Monte Carlo sampling module: assesses the impact of parameter uncertainty on results through repeated sampling; Bayesian estimation module: Calculates the posterior probability distribution of the prediction results based on Bayes' theorem; Sensitivity analysis module: Identifies the input parameters that have the greatest impact on the results using partial differential equations; Confidence Interval Generator: Provides confidence intervals for prediction results based on statistical distribution theory.

5. The method for dynamic monitoring of current transformer error according to claim 1, characterized in that, The state representation module performs the following preprocessing on the input heterogeneous data: Minimum normalization is applied to continuous parameters to restrict their range of values ​​to a specific interval; Sine and cosine transformations are applied to periodic parameters to avoid period jumps. Discrete parameters are processed using one-hot encoding.

6. The method for dynamic monitoring of current transformer error according to claim 1, characterized in that, The steps for quantifying the causal effects of various events on the changes in current transformer error include: Identify the points of error variation in the analysis target and the set of events that may be related to them; By using a multi-world simulation engine, we can obtain predicted error changes under real-world and counterfactual scenarios and create parallel state evolutions for real-world and counterfactual paths. The causal effect of an event on error change is calculated using the average treatment effect formula, which is the difference between the expected value of error change in the actual scenario and the expected value of error change in the counterfactual scenario. To perform robustness verification, the calculation was repeated multiple times under random initial conditions, and the confidence interval of the causal effect was calculated using statistical distribution theory. Z-Scre normalization is applied to the causal effects of all events to convert effect values ​​of different scales and units into comparable uniform dimensions.

7. The method for dynamic monitoring of current transformer error according to claim 1, characterized in that, The steps for real-time monitoring and identification of key triggering events affecting current transformer errors include: Establish a priority queue for causal event monitoring, giving priority to events with high causal effect intensity; The causal effect strength threshold for triggering alarms is dynamically adjusted based on the historical statistical characteristics of the system's operating status using an adaptive threshold algorithm. The sliding time window analysis algorithm determines the size of the time window and the sliding step size, calculates the local causal effect value for the data in each sliding window, and compares the causal effect values ​​of adjacent windows to detect the dynamic changes in causal relationships. The event combination effect evaluation algorithm includes a combination identification algorithm to detect frequently co-occurring event combinations, an interaction effect calculation algorithm to quantify the synergistic or antagonistic effects between events, and an analytic hierarchy process (AHP) algorithm to decompose event combinations into main effects and interaction effects.

8. The method for dynamic monitoring of current transformer error according to claim 7, characterized in that, Sliding time window analysis algorithms include: Determine the time window size and sliding step size; Calculate the local causal effect value for the data within each sliding window; Compare the causal effect values ​​of adjacent windows to detect dynamic changes in causal relationships; When the change exceeds a preset threshold, a re-evaluation is triggered.

9. The method for dynamic monitoring of current transformer error according to claim 1, characterized in that, Also includes: Generate dynamic causal explanation reports, including causal link visualization charts of error changes; Provide error compensation suggestions for different types of events; Construct an error trend prediction model; Build a knowledge base, accumulate experience in causal analysis, and continuously optimize compensation strategies.

10. A dynamic monitoring system for current transformer errors, used to execute the dynamic monitoring method for current transformer errors according to any one of claims 1-9, characterized in that, include: The data acquisition and preprocessing unit is used to acquire timing measurement data of the current transformer and system event logs, and to perform amplitude normalization, standardization and event type encoding conversion. The temporal cause-effect graph construction unit is used to construct a directed weighted graph structure. It represents the temporal causal relationship between system events and error changes through error change point identification algorithm, event sequence normalization algorithm, graph node creation algorithm, time window constraint algorithm, directed edge creation algorithm, edge weight calculation algorithm, and edge pruning algorithm. The counter causal model unit, based on event frequency statistics and time series correlation analysis, identifies a set of potential causal relationship candidates that have passed statistical significance tests through an event counter module, an error change monitoring module, a correlation analysis module, a probabilistic reasoning module, and a causal link screening module. The multi-world simulation engine, based on state space search and conditional reasoning, includes a state representation module, an event intervention module, a state transition calculation unit, and an uncertainty quantification unit. It intervenes in specific events through event operation functions to generate parallel state evolutions of actual and counterfactual paths. The causal effect quantification unit is used to calculate the causal effect of an event on error changes using the average treatment effect formula, perform robustness verification and calculate confidence intervals, and normalize the causal effect. The critical triggering event monitoring unit is used to establish a causal event monitoring priority queue. Through adaptive threshold algorithm and sliding time window analysis algorithm, it monitors and identifies critical triggering events that affect the error of current transformer in real time. The dynamic report generation unit is used to generate causal link visualization charts of error changes, provide error compensation suggestions for different types of events, and build error trend prediction algorithms and knowledge bases to accumulate causal analysis experience.

Images