Intelligent control system of current transformer

Abstract

The invention discloses an intelligent control system for a current transformer, and particularly relates to the technical field of power system monitoring, which comprises a fluctuation identification unit, a redundancy acquisition unit, an error fitting unit, a protection delay unit and a fault response unit, realizes real-time identification and classification of current fluctuation, and has standby path high-speed sampling and fusion signal construction capabilities. The fluctuation trend is judged through error regression, a protection strategy is dynamically set, response is delayed to avoid misoperation, finally, a high-level protection instruction is triggered and a fault event is output when offset overruns or alignment fails, and the data accuracy and the system response stability in the power supply switching process are guaranteed; according to the method, accurate identification of current fluctuation is realized through multi-dimensional feature extraction, data reliability is enhanced by using redundant sampling and error fitting, and by combining delay control and a hierarchical response mechanism, the risk of misoperation is effectively reduced, the fault judgment accuracy and the system protection stability are improved, and the method is suitable for a complex power supply switching scene.

Description

Technical Field

[0001] The present invention relates to the technical field of power system monitoring, and more specifically, to a current transformer intelligent control system. Background Art

[0002] As a key measurement and protection element in power systems, current transformers are widely used in real-time current detection, state perception, and protection control. Under normal operating conditions, traditional transformers in conjunction with protection units can continuously monitor steady-state current parameters and trigger protection responses when thresholds are exceeded or abnormalities occur.

[0003] However, during the backup power switching phase of data centers or precision industrial equipment, the power switching logic itself has problems such as relay intermittent, phase drift, frequency disturbance, etc., which will cause slight mutations or fluctuations in the current signal in a very short time. Such fluctuations usually have the following characteristics: short duration (such as 5-15 milliseconds); medium disturbance amplitude and tolerance; unpredictable trend, which may not be a real fault. In the above scenario, the existing technology mainly has the following problems: It is impossible to effectively distinguish between tolerable switching fluctuations and real fault signals. Most traditional protection strategies adopt a fixed threshold + fixed response method. When facing transient disturbances, there is a lack of trend judgment and hierarchical processing mechanism, which makes it very easy for false operations or premature triggering of protection responses to occur.

[0004] There is a lack of dynamic tracking and error fitting mechanism for current fluctuation trends. Some current technologies introduce oversampling or redundant acquisition methods, but no stable fitting judgment model has been established, making it difficult to dynamically adjust the protection strategy, resulting in a "rigid response" of the system.

[0005] The protection response behavior lacks buffering and linkage mechanisms, and the execution of control instructions is too rigid. The protection response in the existing technology is basically "judgment established → immediate action", lacking the "state buffer - delayed confirmation - trend linkage" mechanism, and lacks fault tolerance in complex fluctuation scenarios.

[0006] There are still deficiencies in the collaborative judgment of multi-source data, and it is impossible to achieve feedforward control and event structured management. Therefore, the present invention proposes a current transformer intelligent control system to solve the above problems. Summary of the invention

[0007] To achieve the above object, the present invention provides the following technical solutions: A current transformer intelligent control system, comprising: The fluctuation identification unit is used to obtain the current signal during the switching period through real-time sampling, extract the continuous change amplitude, rising edge slope, phase difference change and frequency mutation value from the current signal based on the predefined instantaneous change identification standard, and complete the feature identification and classification of abnormal fluctuation behavior according to data statistical rules; The redundant acquisition unit, after the identification result belongs to the tolerable switching fluctuation type, immediately triggers the backup sampling path and enables the high-speed sampling instruction to obtain the redundant current information flow, performs time axis comparison on the main path data, and constructs the fusion signal result on the premise of satisfying the timing continuity, which is used to replace the distortion or drift segment in the main signal; The error fitting unit continuously compares the fusion signal with the preset reference data model, performs fitting regression on the error deviation trend of the current fluctuation state, sets the protection trigger threshold and response time according to the fitting result, and makes real-time corrections to the main control strategy; The protection delay unit automatically delays the protection execution time when it determines that the fluctuation state is within the acceptable range, and outputs a fusion signal as an effective monitoring benchmark. At the same time, it suspends high-priority protection response instructions to prevent false operations caused by transient errors and improve the power supply continuity during switching. The fault response unit immediately executes the emergency response process, generates a high-level protection instruction and interrupts the signal output process when the offset between the fusion signal and the reference data model exceeds the preset limit, or the redundant sampling path cannot complete the effective time alignment, and sends a fault event identification to the external system for protection processing and log recording of power switching control.

[0008] In a preferred embodiment, the fluctuation identification unit includes an abnormal current fluctuation screening structure, which is used to set a multi-threshold condition judgment mechanism based on the extraction of continuous change amplitude, rising edge slope, phase difference change and frequency mutation value, and to combine and operate different characteristic parameters in a time-aligned manner to construct a high-dimensional characteristic behavior discrimination space. Through the set three-value hierarchical decision logic, the detection results are marked as stable fluctuation, critical fluctuation or severe abnormality. In the time window slip judgment process, whether the fitting coefficient between the average amplitude change and the rising rate of three adjacent sampling periods is lower than 0.1 is used as the first judgment factor, and whether the instantaneous frequency change is within 5 Hz is used as the second judgment factor. When both conditions are met at the same time, the current state is judged to be a critical fluctuation. The critical fluctuation type is identified as a tolerable switching fluctuation type, which is used as the basis for starting the subsequent backup sampling path, so that the working behavior of the redundant acquisition unit has a traceable judgment basis based on the feature recognition results, thereby ensuring the continuity of the technical chain between fluctuation judgment and data compensation.

[0009] In a preferred embodiment, the redundant acquisition unit establishes a data mirror relationship by operating the dual-path current signal channel in parallel, wherein the setting error between the main path sampling period and the backup path sampling period does not exceed ten microseconds, so as to ensure that the time alignment accuracy meets the data fusion requirements; The backup path sampling value needs to go through a preliminary noise elimination process, where the elimination method includes one of the three methods: maximum difference value method, root mean square limit method, and adaptive judgment based on sliding window filtering. The specific method is automatically switched according to the current data fluctuation type and is not fixedly bound for use; Under the condition that the sampling values ​​satisfy the redundant timing structure, an interleaved data stream is established by delaying one cycle to ensure the continuity of the redundant data and to ensure that it does not cause logical interference with the main path. At the same time, all redundant data will be brought into the phase reconstruction operation process to participate in the construction of the error fitting unit behavior.

[0010] In a preferred embodiment, the error fitting unit comprises a dynamic regression correction structure, wherein the regression behavior is analyzed based on the mean deviation trend of the fused signal from the reference model within the current five sampling periods, while considering the consistency of the deviation direction and the stability of the duration; The regression strategy adopts one of two methods: interval minimum variance fitting method and exponential weighted moving average method. The former is used when the current fluctuation intensity is greater than ten percent, and the latter is used when the fluctuation duration is less than ten milliseconds. The offset is converted into a trigger adjustment amplitude factor, whose value range is from 0.05 to 0.15, and is used to correct the protection trigger threshold; When the fusion signal has a trend-reverse fluctuation behavior, the fitting strategy automatically enters the hysteresis weight adjustment logic, so that the protection response maintains adjustable continuity within the stable response range and avoids frequent oscillating operations.

[0011] In a preferred embodiment, the error fitting unit further includes a trend segment tracking structure for separating the influence of high-frequency disturbances, analyzing the behavior trend by fitting residuals in each detection cycle, and entering the conservative correction phase of the fitting trend when the residual is less than a reference value of 0.03 for three consecutive cycles, in which no rapid threshold regression adjustment is performed, and the protection strategy is updated only based on the historical cumulative average error; When the residual reverses direction twice within five cycles, the automatic output signal does not participate in the threshold correction and only maintains the current delay behavior.

[0012] In a preferred embodiment, the protection delay unit includes a behavior priority conversion structure, and when the fusion signal is classified as a critical fluctuation type, the protection execution time interval is actively delayed, and the delay value is not less than five milliseconds and not more than twenty milliseconds. The specific delay time is obtained by linear mapping based on the adjustment amplitude factor output by the error fitting unit; During the delay period, all signal outputs are marked as medium risk, and high-level interrupt control commands are prohibited from being sent; after the delay, if the regression value of the error fitting unit remains stable, the delay will continue; otherwise, the protection action call behavior in the fault response unit will be executed; the priority conversion structure ensures that before the abnormal state is accurately verified, the system response has a clear waiting mechanism to minimize the probability of false operation.

[0013] In a preferred embodiment, the protection delay unit generates a delay behavior state label after entering the delay state, and the label includes the delay time interval, the current fusion signal fitting residual value, the amplitude change trend parameter and the fitting regression level information, which is embedded in the output signal stream in real time and transmitted to the fault response unit; The delayed behavior status label is automatically updated every two sampling cycles to assist the fault response unit in executing the response level determination process. When the fitting regression level in the status label shows a deteriorating trend or the residual value continues to expand, the fault response unit loads the response judgment condition in advance and enters the candidate preparation state; The delay behavior status label is also used as one of the input fields for generating the fault event identification label. Its content is written into the event identification structure after the fault response behavior is triggered to support the subsequent external fault trend analysis and multi-label clustering process. Through the generation and dynamic update of the delay behavior status label, a data sharing mechanism is established between the protection delay unit and the fault response unit to realize the feedforward judgment of the response path and cross-unit linkage.

[0014] In a preferred embodiment, the fault response unit introduces a delayed behavior state label as a preprocessing information source for response judgment before executing the emergency response process, wherein the fitting regression level and residual trend in the state label are used as one of the judgment factors, and the offset degree, duration and fitting trend change between the current fusion signal and the reference model are combined to jointly determine the response behavior level; The response behavior is divided into three categories: interrupt output, signal bypass and data freeze. If the residual value in the delay behavior state label increases continuously and reaches five cycles, and the fusion signal deviation degree is greater than 15% for three consecutive cycles, the interrupt output is triggered; If the tag trend does not deteriorate significantly but the fluctuation lasts less than ten milliseconds, the data freeze strategy is executed; If the fitting trend shows nonlinear growth and the status label level is medium risk, the signal bypass logic is executed to redirect the current data flow to the redundant path output channel; this response mechanism introduces delayed behavior status labels as linkage conditions to improve the stability and reliability of response trigger judgment.

[0015] In a preferred embodiment, the fault response unit generates a fault event identification tag when triggering a protection behavior. The tag structure includes a fault type code, a fluctuation behavior tag, a protection action record, a fusion signal fitting trajectory information, and a delay behavior status tag field. It is encapsulated and output synchronously in a standard communication data frame format. Each event identification tag has a unique number for fault tracing and historical archiving. When there are similar fitting trends, similar delay behavior status fields, and different fluctuation starting points in multiple event identification tags, the fault trend clustering logic is triggered, and an early warning report is sent to the superior platform.

[0016] Technical effects and advantages of the present invention: The present invention realizes real-time sampling and high-dimensional feature extraction of current signals during power switching through a fluctuation recognition unit. The extracted continuous change amplitude, rising edge slope, phase difference change, frequency mutation value and other indicators are combined with predefined instantaneous change recognition standards and data statistical rules to identify abnormal behavior and classify them, thereby having the ability to identify the start of fluctuations and distinguish normal disturbances from potential abnormalities. Compared with the traditional judgment method based only on the change of current amplitude, the present invention can carefully perceive the short-term disturbance state, accurately identify tolerable fluctuations, and avoid misjudging non-fault fluctuations during power switching as faults.

[0017] The present invention compares, compensates and fuses the main path signal through the backup sampling path startup mechanism of the redundant acquisition unit, ensuring that a time-series continuous and highly reliable fusion signal can be constructed under abnormal interference conditions such as transient drift and inductive distortion of the main channel signal; at the same time, the error fitting unit performs dynamic regression analysis on the error offset trend between the fusion signal and the reference data model, and sets the protection trigger threshold and response time according to the fitting result, thereby effectively improving the data stability and judgment accuracy of the system before protection execution, and providing a true and reliable current state basis for subsequent control logic.

[0018] The present invention uses a linkage mechanism between the protection delay unit and the fault response unit to automatically delay the execution of the protection instruction when the fluctuation is recognized to be in an acceptable state, suspend the high-priority protection response, and wait until the fusion signal fitting trend deteriorates or the offset exceeds the preset limit before the fault response unit generates a high-level protection instruction and executes the interruption or interruption processing. The response process introduces clear delay buffering, trend verification and hierarchical protection logic, effectively reducing the probability of false triggering due to instantaneous disturbances or sampling anomalies, while ensuring the timeliness of fault response and the stability of system protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to facilitate understanding by those skilled in the art, the present invention is further described below in conjunction with the accompanying drawings; Figure 1It is a schematic diagram of a current transformer intelligent control system in the present invention; Figure 2 This is a working principle diagram of the fluctuation identification unit in the present invention; Figure 3 This is a working principle diagram of the fault response unit in the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Reference Figure 1-3 The following embodiments are obtained: Example

[0022] A current transformer intelligent control system, comprising: The fluctuation identification unit is used to obtain the current signal during the switching period through real-time sampling. Based on the predefined instantaneous change identification standard, the continuous change amplitude, rising edge slope, phase difference change and frequency mutation value are extracted from the current signal, and the feature identification and classification of abnormal fluctuation behavior are completed according to the data statistical rules. This unit is the "perception entrance" of the entire system, and its goal is to perform high-precision classification and identification of abnormal current fluctuations that may occur during the backup power supply switching. Traditional current transformer systems only use current amplitude as a judgment standard, which is prone to false alarms or omissions during the power switching period. The fluctuation identification unit constructs a multi-dimensional feature space by extracting richer signal features (such as slope, phase difference, frequency mutation), so as to accurately determine whether it is a normal switching disturbance or a real fault abnormality, providing a classification basis for all subsequent behaviors.

[0023] The redundant acquisition unit, after the identification result belongs to the tolerable switching fluctuation type, immediately triggers the backup sampling path, enables high-speed sampling instructions to obtain redundant current information flow, performs time axis comparison on the main path data, and builds a fusion signal result on the premise of satisfying the timing continuity, which is used to replace the distortion or drift segment in the main signal; after the fluctuation identification confirms that the current fluctuation belongs to the tolerable type, the system enters the data fault tolerance stage. The redundant acquisition unit is responsible for the verification and compensation of the main path sampling signal. By starting the backup sampling path, performing high-speed data acquisition and timing comparison, and building a fusion signal, the signal distortion problem caused by phase drift, induction saturation, external interference, etc. in the main channel is solved, ensuring that the system judgment is based on real and reliable data and improving the overall judgment robustness.

[0024] The error fitting unit continuously compares the fusion signal with the preset reference data model, fits and regresses the error deviation trend of the current fluctuation state, sets the protection trigger threshold and response time according to the fitting results, and makes real-time corrections to the main control strategy; the dynamic changes of the fusion signal reflect the evolution of the system state, and the error fitting unit tracks the fluctuation deviation trend in real time by comparing with the reference model, and performs mathematical fitting modeling. The fitting results are not only used to adjust the triggering conditions (threshold, delay) of the protection action, but also bear the key function of judging the stability trend of the system and whether it is in the process of deterioration. It is the core logic center for converting data processing into control decisions.

[0025] The protection delay unit automatically delays the execution of protection when it determines that the fluctuation state is within an acceptable range, and outputs a fusion signal as an effective monitoring benchmark. At the same time, it suspends high-priority protection response instructions to prevent false operations due to transient errors and improve the continuity of power supply during switching. To avoid short-term fluctuations triggering unnecessary protection actions, the protection delay unit provides a protection response buffer mechanism with an intelligent judgment basis. According to the output trend of the error fitting, the timing and strategy of protection execution are dynamically adjusted, the control response is delayed when the system is still tolerable, and the fusion signal is continuously used as a monitoring benchmark. This design significantly reduces false operations caused by transient disturbances and improves the system's anti-interference ability and control flexibility in actual operation.

[0026] The fault response unit immediately executes the emergency response process, generates a high-level protection instruction, interrupts the signal output process, and sends a fault event identification to the external system for protection processing and log recording of power switching control when the offset between the fusion signal and the reference data model exceeds the preset limit, or the redundant sampling path cannot complete effective time alignment. When the system determines that the fluctuation deviates from the trend seriously or the compensation fails, it must be dealt with quickly. The fault response unit introduces a high-priority judgment mechanism to make protection decisions immediately when redundant data fails or the offset exceeds the limit, including output interruption, fault recording, external reporting, etc. This unit ensures that the current transformer can quickly protect the safety of key equipment and systems under abnormal circumstances, and is the "execution terminal" of the entire protection chain.

[0027] The fluctuation identification unit includes an abnormal current fluctuation screening structure, which is used to set a multi-threshold condition judgment mechanism based on the extraction of continuous change amplitude, rising edge slope, phase difference change and frequency mutation value. Different characteristic parameters are combined and operated in a time-aligned manner to construct a high-dimensional characteristic behavior discrimination space. Through the set three-value hierarchical decision logic, the detection results are marked as stable fluctuation, critical fluctuation or severe abnormality. In the time window slip judgment process, whether the fitting coefficient between the average amplitude change and the rising rate of three adjacent sampling periods is lower than 0.1 is used as the first judgment factor, and whether the instantaneous frequency change is within 5 Hz is used as the second judgment factor. When both conditions are met at the same time, the current state is judged to be a critical fluctuation. The critical fluctuation type is identified as a tolerable switching fluctuation type, which is used as the basis for starting the subsequent backup sampling path, so that the working behavior of the redundant acquisition unit has a traceable judgment basis based on the feature recognition results, thereby ensuring the continuity of the technical chain between fluctuation judgment and data compensation.

[0028] The fluctuation identification unit is used to continuously sample the input current signal in real time during the backup power supply switching, and extract multiple fluctuation characteristic parameters, including: continuous change amplitude (i.e., the absolute mean of the difference between the current values ​​of adjacent sampling points, used to measure the intensity of fluctuations), rising edge slope (indicates the growth rate of the current signal in the rising section, reflecting the instantaneous transition trend), phase difference change (indicates the change amplitude of the phase difference between the current signal and the reference clock or reference power supply within a certain period of time, used to determine whether the synchronization is unstable), and frequency mutation value (by analyzing the time interval between zero crossing points to reverse the instantaneous frequency change, to identify whether the system has short-term switching or harmonic disturbances). After the above characteristic parameters are collected, the feature identification and classification attribution process of abnormal fluctuation behavior is entered. This process is judged based on pre-set data statistical rules. The statistical rules may include multi-factor condition combination judgment, interval classification threshold setting, and feature trend fitting within the sliding window.

[0029] In order to improve the recognition accuracy, a time window sliding judgment mechanism is introduced in the recognition process, that is, several characteristic parameters of the current current signal are analyzed within a dynamic sliding time interval. For example, within three consecutive sampling periods (each group can contain a fixed number of sampling points, such as 20 points per group), the average amplitude change is calculated, that is, the average amplitude change between the maximum and minimum current values ​​in each group of samples; then the rising rate is calculated, that is, the slope of the fitting line of the current value of the rising section of the signal to the time point, and the fitting coefficient is calculated for the two. The "fitting coefficient" here refers to the slope fitting calculated by the least squares method, which is a quantitative indicator to characterize the degree of linear correlation between the two. If the fitting coefficient is lower than zero point one (indicating that the linear correlation between the two is very weak and there is no obvious growth trend in the fluctuation behavior), it is considered that the first judgment factor meets the conditions.

[0030] The system then evaluates the instantaneous frequency change value and determines the frequency change trend through the timing difference of the zero crossing point of the sampling signal. If the fluctuation frequency remains within the range of five hertz (for example, between 48.5Hz and 51.5Hz), it is considered that the system main frequency has not been greatly disturbed and meets the second judgment factor condition. When both the first judgment factor and the second judgment factor are met, the system attributes the period identification result to critical fluctuation.

[0031] The definition of critical fluctuation is: the current signal shows a certain amplitude and fluctuation trend, but overall it does not exceed the main control threshold and does not show instability characteristics. It is a typical tolerable disturbance phenomenon, which usually occurs when the power supply enters the UPS backup path from the main power due to short-term signal disturbances caused by relay action, instantaneous electromagnetic interference, etc. Therefore, the critical fluctuation type is identified by the system as a tolerable switching fluctuation type as the basis for entering the subsequent data compensation and sampling switching process.

[0032] When the current recognition result output by the fluctuation recognition unit is a critical fluctuation type, the backup sampling path startup instruction in the redundant acquisition unit is immediately triggered, and then high-speed sampling is started to ensure that data compensation can be performed on the basis of accurate judgment. This method enables the working behavior of the redundant acquisition unit to have a clear and traceable feature recognition source, and builds a technical closed-loop chain from "feature acquisition → judgment classification → data compensation", ensuring that there is no logical jump or fuzzy judgment between different processing logics, and improving the reliability and robustness of system judgment.

[0033] The redundant acquisition unit builds a data mirror relationship by running the dual-path current signal channel in parallel, where the setting error between the main path sampling period and the backup path sampling period does not exceed ten microseconds to ensure that the time alignment accuracy meets the data fusion requirements; The backup path sampling value needs to go through a preliminary noise elimination process, where the elimination method includes one of the three methods: maximum difference value method, root mean square limit method, and adaptive judgment based on sliding window filtering. The specific method is automatically switched according to the current data fluctuation type and is not fixedly bound for use; Under the condition that the sampling values ​​satisfy the redundant timing structure, an interleaved data stream is established by delaying one cycle to ensure the continuity of the redundant data and to ensure that it does not cause logical interference with the main path. At the same time, all redundant data will be brought into the phase reconstruction operation process to participate in the construction of the error fitting unit behavior.

[0034] In order to ensure the stability and reliability of current monitoring data during power switching, the redundant acquisition unit uses a parallel operation mode of dual-path current signal channels to build a data mirror structure.

[0035] Definition and purpose of dual-path current signal channel: The so-called "dual-path channel" means that the main path and the backup path independently receive the same current source output signal, and sample, store and process it through their respective sampling units. Among them: The main path is responsible for routine sampling, that is, the system standard sampling process, with a moderate response speed, and is used for normal operation data output.

[0036] The backup path is activated after identifying the "tolerable switching fluctuation type" and enters a high-speed sampling state. Its task is to perform higher frequency and higher precision compensation acquisition of the detailed behavior during the fluctuation.

[0037] In order to ensure that the sampled data between the main path and the backup path can be fused, strict time alignment accuracy requirements must be met. In the present invention, the error between the main path sampling period and the backup path sampling period is controlled to be no more than ten microseconds, that is, the time sampling interval difference between the two channels shall not be greater than 10μs. This setting is to ensure that in the subsequent fusion stage, the redundant data and the main data can be compared one-to-one on the time axis to prevent data dislocation, interleaving or redundant overflow, and ensure the accuracy of the fusion signal construction.

[0038] Definition of noise rejection mechanism and rejection method: In order to improve the effectiveness of the backup path sampling data, the sampled value needs to undergo preliminary noise rejection before entering the data fusion process. Noise rejection refers to the exclusion of non-real current change data introduced by sensors, electromagnetic interference or sampling deviation. The rejection method includes the following three processing mechanisms: Maximum difference value method: Based on the difference between adjacent sampling points, a rejection threshold is set. All samples with a difference greater than the threshold are considered as abnormal points and rejected. Root mean square limiting method: Calculate the root mean square value (RMS) of the sampling sequence, and then set the upper and lower limit amplitude ranges. Sampling points beyond the range are regarded as noise. Adaptive judgment based on sliding window filtering: Within the sliding time window (such as 5-10 o'clock), the sample mean and standard deviation are dynamically calculated, and an adaptive algorithm is used to determine whether the sampling point deviates from the statistical concentration value, and then determine whether to remove it.

[0039] The three methods are not fixed to one, but the elimination method is automatically switched according to the current data fluctuation type (i.e., the fluctuation feature level output by the fluctuation identification unit, such as critical fluctuation or stable fluctuation). For example, in scenes with high fluctuation intensity, the sliding window filtering method tends to be used; while in scenes with periodic disturbances, the maximum difference value method may be more appropriate. This mechanism enhances the flexibility and scene adaptability of noise processing, avoiding "over-elimination" or "wrong retention".

[0040] Interleaved data stream construction and fusion mechanism: After the backup path sampling value is eliminated and deemed valid, the next step is to enter the redundant timing structure comparison stage. In this stage, the interleaved data structure is established by delaying one cycle, that is, the backup path sampling point lags slightly behind the main path by one cycle (for example, if the sampling cycle is 10μs, the backup path is delayed by a 10μs cycle), forming a cross timing structure.

[0041] This method has two effects: Avoid complete overlap of processing logic between the primary and backup paths, which may lead to resource conflicts. When signal distortion occurs, the backup path can provide front and back redundant points for interpolation compensation to improve the stability of the fusion result.

[0042] The output structure of the interleaved data stream will be sent to the next step of the fusion signal construction process. In this process, the sampling sequence in the backup path not only plays a role in data compensation, but also serves as a reference for comparing the sampling accuracy and integrity of the main path, ensuring that the data fusion has temporal continuity and credibility.

[0043] Phase reconstruction operation process description: Redundant data is also uniformly brought into the phase reconstruction operation process. The core of this process is to analyze the relative phase drift of the backup sampling value, perform fitting correction through the reference phase model (which can be obtained from the historical stable cycle or reference frequency), and output an accurate reference signal that can be used for subsequent error fitting units. Phase reconstruction is to deal with data errors caused by power supply phase drift during switching in the main path, so that the final fusion signal is consistent in three dimensions: time, amplitude and phase, providing a real and reliable current state input for subsequent protection actions.

[0044] The error fitting unit contains a dynamic regression correction structure, in which the regression behavior is analyzed based on the mean deviation trend of the fusion signal from the reference model in the current five sampling periods, while considering the consistency of the deviation direction and the stability of the duration; The regression strategy adopts one of two methods: interval minimum variance fitting method and exponential weighted moving average method. The former is used when the current fluctuation intensity is greater than ten percent, and the latter is used when the fluctuation duration is less than ten milliseconds. The offset is converted into a trigger adjustment amplitude factor, whose value range is from 0.05 to 0.15, and is used to correct the protection trigger threshold; When the fusion signal has a trend-reverse fluctuation behavior, the fitting strategy automatically enters the hysteresis weight adjustment logic, so that the protection response maintains adjustable continuity within the stable response range and avoids frequent oscillating operations.

[0045] The error fitting unit is responsible for comparing the fusion signal generated by the redundant acquisition unit with the reference data model preset by the system in real time, and through trend analysis and dynamic regression technology, it completes the judgment of the degree of fluctuation deviation and the adaptive correction of the protection strategy. To achieve this function, the error fitting unit contains a dynamic regression correction structure, the core of which is to calculate whether the current fluctuation is in a continuous deviation state through regression modeling, and adjust the protection control parameters accordingly.

[0046] Definition and purpose of dynamic regression correction structure: This structure uses the fusion signal as the input data source, and compares it with the expected current value in the system reference model one by one within the current five sampling cycles, that is, within the time window composed of the last five groups of equally spaced sampling points, and calculates its mean shift trend to identify whether the current current signal is undergoing a systematic shift. The judgment of the shift trend not only depends on the numerical difference, but also needs to meet two trend conditions at the same time: Offset direction consistency: that is, the offset values ​​within multiple sampling periods should have the same change direction (continuous increase or decrease) to avoid occasional fluctuations being misjudged as continuous offsets; Duration stability: The deviation trend must persist for more than two complete sampling periods (for example, at least 3 consecutive periods in 5 periods show the same direction of change) to confirm that the fluctuation is statistically persistent. After the above conditions are met, the specific regression strategy selection process begins.

[0047] Regression strategy selection logic and algorithm description: After the error fitting unit passes the trend judgment, it will select one of the following two regression strategies for calculation according to the specific characteristics of the current fluctuation: Interval minimum variance fitting method: This method is applicable to the case where the current fluctuation intensity is greater than 10%, that is, the maximum deviation of the fusion signal from the reference model exceeds 10% of its rated value. By performing the least squares fitting on the current five-cycle sampling value, a trend regression line is generated to reflect the current fluctuation direction and rate, and has the ability to respond quickly to large fluctuations.

[0048] Exponentially weighted moving average method: When the fluctuation duration is less than ten milliseconds, that is, the fluctuation occurs quickly but for a short period of time, the system uses this method to mitigate the calculation. This method achieves a smooth response to sudden changes by giving higher weights to recent sampling values, avoiding misjudgments caused by data anomalies in a short period.

[0049] Once the regression strategy is selected, the current fitting offset will be output. Interval minimum variance fitting is a trend modeling method based on the minimum residual sum of squares. Its core goal is to find a straight line or smooth curve within a fixed time window so that the sum of square errors between the curve and the fusion signal is minimized. The data processing flow is as follows: arrange the fusion signal values ​​of the current five consecutive sampling cycles in chronological order; use the corresponding reference model values ​​as the comparison baseline; use time as the independent variable and the current difference as the dependent variable, and find a trend line through the principle of minimum square error sum; the slope of the trend line represents the direction and speed of the current signal offset.

[0050] Exponentially weighted moving average (EWMA) is a strategy that emphasizes the weight of recent data and smoothes sudden disturbances. The basic idea is that the newer the data point, the greater its proportion in the average calculation, thereby enhancing the system's real-time sensitivity to the current state and suppressing occasional anomalies. The data processing flow is: extract the fusion signal value within the current five sampling cycles; set the initial weight distribution rule (for example, the last cycle has the largest weight and decreases forward); sum all data according to the weight and subtract it from the reference model average; finally calculate the smoothing offset of the current window.

[0051] The offset is converted into a control parameter, the trigger adjustment amplitude factor. The offset in the fitting result is the mean difference between the current fusion signal and the reference model. After the trend is determined to be established, the offset will be converted into a control parameter, such as through a preset mapping relationship, called the trigger adjustment amplitude factor. The value range of this factor is limited to between 0.05 and 0.15. The larger the value, the closer the current system is to the critical value that needs to respond.

[0052] This factor is used to adjust the protection trigger threshold in the protection delay unit. For example, when the fitting offset is positive and exceeds the upper limit range, the original protection current threshold is adjusted downward by a certain percentage to speed up the system response; if it is a negative offset, the protection response is delayed to improve the system tolerance. Through this dynamic control method, the system can better adapt to the stability requirements under different loads or switching conditions.

[0053] Trend reverse fluctuation and hysteresis weight logic: In actual operation, the fusion signal fitting trend may reverse direction, that is, the fitting curve changes from rising to falling or vice versa, indicating that the system state has changed unstable. To avoid control oscillation caused by frequent switching, the error fitting unit will automatically start the hysteresis weight adjustment logic in this case: In the weight adjustment logic, a hysteresis factor within an additional time window is applied to the fitting algorithm; the values ​​of recent cycles are given lower weights, and the historical stable cycle values ​​dominate; through the preset hysteresis smoothing strategy, the protection response is kept within the stable response range, which can effectively avoid the oscillation behavior caused by frequent switching protection logic and improve the system response continuity and stability of the current transformer in complex transient scenarios.

[0054] The error fitting unit also includes a trend segment tracking structure, which is used to separate the influence of high-frequency disturbances. In each detection cycle, the behavior trend is analyzed by fitting residuals. When the residual is less than the reference value of 0.03 for three consecutive cycles, the fitting trend conservative correction stage is entered. In this stage, no rapid threshold regression adjustment is performed, and the protection strategy is updated only based on the historical cumulative average error. When the residual reverses direction twice within five cycles, the automatic output signal does not participate in the threshold correction and only maintains the current delay behavior.

[0055] In the embodiment of the present invention, the error fitting unit includes not only a dynamic regression correction structure but also a trend segment tracking structure. This structure is used to enhance the ability to distinguish short-term disturbances from real trends under complex or unstable current conditions, especially to suppress and adjust strategies for residual fluctuations caused by high-frequency disturbances.

[0056] "Trend segmented tracking structure" means: within a fixed monitoring window, the fitting residuals between the fusion signal and the reference data model are statistically analyzed period by period, and the stability of the signal fluctuation behavior is judged based on the continuity and change direction of the residuals, and then a decision is made whether to perform regression adjustment or enter the conservative stage of the protection strategy.

[0057] Fitting residual: refers to the difference between the fusion signal and the regression fitting curve at each sampling point, which is used to measure the degree to which the current data point deviates from the trend line; High-frequency disturbance influence: refers to a small oscillation signal that changes rapidly in a short period of time, usually caused by electromagnetic interference, relay rebound, power switching edge state, etc., and has nothing to do with the actual change of system current; Detection cycle: It is the sliding analysis cycle defined by the system, usually a number of consecutive sampling cycles (for example, each cycle contains 5-10 groups of sampling points); Behavioral trend: refers to the overall deviation direction determined based on the change in the direction of continuous residuals.

[0058] The system normalizes the fitting residuals of all sampling points in each detection cycle, calculates their absolute values ​​and compares them with a fixed reference value. The reference value is set to 0.03, which means that residuals within ±0.03 current units (such as amperes) can be considered "stable".

[0059] When the fitting residual is less than the reference value for three consecutive cycles, it means that the current fusion signal fluctuates very little. At this time, the system judges that the current state has a stable trend and there is no sudden abnormality. The system will automatically enter the conservative correction stage of the fitting trend. The characteristics of the conservative correction stage are: in this stage, the error fitting unit no longer performs fast fitting regression, that is, it does not recalculate the trend line slope, nor re-evaluate the threshold response factor, but only uses the historical cumulative average error for control judgment.

[0060] "Historical cumulative average error" means: taking a weighted average of the residual means in multiple previous normal cycles as a control reference. This method is more conservative and more tolerant than the fast fitting response, and is suitable for improving system robustness in stable scenarios.

[0061] In another case, the trend segment tracking structure also pays attention to the direction of change of the fitting residual, that is, whether the residual increases positively or decreases negatively. The system identifies whether the trend change has reversed by comparing period by period: when the fitting residual reverses direction more than twice in five consecutive detection periods (for example, the residual changes from positive to negative, and then from negative to positive); the system determines that: the current fluctuation trend is uncertain, oscillatory, or affected by high-frequency interference; in order to avoid triggering false protection actions when the trend is not clear, the system does not perform any threshold correction behavior. In this case, the system will automatically maintain the current protection delay behavior and will not increase or decrease the protection level. This method can prevent frequent switching of protection strategies and avoid protection misoperation or protection strategy fatigue caused by jitter.

[0062] Assume that when the backup power supply is switched, the system fusion signal fluctuates slightly for 5 consecutive cycles, and the residual value is always less than 0.02. At this time, the system enters the conservative correction stage, and only uses the historical average value for judgment, without recalculation. In another scenario, in 5 cycles, the system residual changes in the direction of "up → down → up" repeatedly. Although the residual amplitude does not exceed the standard, the system stops updating the threshold because the trend reverses twice, and only maintains the original delay time, waiting for the trend to become clear before deciding to respond.

[0063] The protection delay unit includes a behavior priority conversion structure. When the fusion signal is classified as a critical fluctuation type, the protection execution time interval is actively delayed. The delay value is not less than five milliseconds and not more than twenty milliseconds. The specific delay time is obtained by linear mapping based on the adjustment amplitude factor output by the error fitting unit. During the delay period, all signal outputs are marked as medium risk, and high-level interrupt control commands are prohibited from being sent; after the delay, if the regression value of the error fitting unit remains stable, the delay will continue; otherwise, the protection action call behavior in the fault response unit will be executed; the priority conversion structure ensures that before the abnormal state is accurately verified, the system response has a clear waiting mechanism to minimize the probability of false operation.

[0064] In an embodiment of the present invention, the protection delay unit is provided with a behavior priority conversion structure, which is used to intelligently judge and grade the execution timing and response level of the protection action according to the severity of the current current fluctuation state and the system identification level, so as to avoid unnecessary protection responses due to misjudgment or transient disturbances.

[0065] The "behavior priority conversion structure" refers to a control mechanism that divides the protection response behavior into multiple priorities according to the current identification results of the system in the current transformer control logic, and selects whether and when the protection action is executed through the priority logic. Its design purpose is to introduce a response buffer strategy so that the system can actively delay the response time when the judgment is not sufficient or the state is not stable, thereby avoiding false operations caused by short-term abnormalities or false signals.

[0066] The fusion signal is classified as a critical fluctuation type response logic: In actual operation, the fusion signal is the output result of the current data of the main path and the backup path after timing comparison, error compensation, and phase calibration. This signal is the most reliable reflection of the current state of the system.

[0067] If the fusion signal is judged by the fluctuation recognition unit to belong to the "critical fluctuation type", that is, the fluctuation amplitude is within the tolerance range, the frequency change has no drastic mutation, and the fitting trend has not deteriorated, but there is still potential for change, the system will activate the behavior priority conversion structure and enter the response delay process.

[0068] Delay mechanism and control logic of protection execution time: When the behavior priority conversion structure is activated, the system will immediately and actively delay the protection execution time interval, that is, delay the execution of the protection instructions originally planned to be executed in the current cycle for a period of time to observe whether there is a trend of deterioration.

[0069] Time delay range setting: for example, the delay time is not less than five milliseconds and not more than twenty milliseconds; the specific delay time is linearly calculated based on the adjustment amplitude factor output by the error fitting unit; the so-called "adjustment amplitude factor" refers to the dynamic coefficient generated by the error fitting unit based on the current fusion signal and the reference model offset degree and the trend fitting result, and the range is usually set between 0.05 and 0.15; the delay time is proportional to the factor: the smaller the factor, the shorter the delay; the larger the factor, the higher the trend risk and the longer the delay time. This mapping method ensures that the system has adaptive response capabilities to fluctuations of different severity.

[0070] Medium-risk marking and command suppression mechanism: In the delayed phase, to ensure that the system's external control devices and superior systems do not misunderstand the current status, the behavior priority conversion structure temporarily marks the output status: all signal outputs are marked as "medium risk", which informs the peripheral system that the current status has not been confirmed but requires attention; during this phase, the system is prohibited from sending high-level interrupt control commands, such as circuit breaker tripping, master device switching, etc.; this suppression measure prevents external execution of false actions while retaining a window for the system to continue observing and correcting.

[0071] The judgment behavior after the delay expires is connected with the fault response: after the delay time interval ends, the system makes a judgment based on the latest output result of the error fitting unit: if the fitting regression value remains stable, that is, the residual change has not significantly expanded, and the fluctuation direction has not continued to deteriorate; then the delay state is automatically renewed, the delay time is reset, and the observation state is entered again; if the fitting value shows a trend of deterioration or the offset exceeds the limit, it is considered that the abnormality has been confirmed, the system exits the delay state, and immediately enters the fault response unit execution stage; the fault response unit will call the corresponding protection action according to the event level, such as output interruption, signal bypass or data freezing.

[0072] Through this judgment logic, the system realizes dynamic identification of "recoverable state" and "response-required state", thereby postponing the response strategy until sufficient judgment basis is established, reducing the probability of false response and improving the accuracy of system judgment.

[0073] For example: During a UPS power switching process, the system fusion signal detected that the current showed a medium-intensity jump within two cycles, which was judged as a critical fluctuation type by the fluctuation identification unit. The adjustment amplitude factor output by the error fitting unit is 0.10, and the corresponding delay time is 15 milliseconds. The system enters the delay state, prohibits the output of the power-off command, and marks the state output as medium risk. During the delay period, the fitting trend has not changed, and the system automatically renews for 5 milliseconds. After re-evaluation, it is found that the residual has increased, and it is judged that the risk has increased. Therefore, the system exits the delay state, enters the fault response logic, and executes the signal bypass action.

[0074] After entering the delayed state, the protection delay unit generates a delay behavior state label, which contains the delay time interval, the current fusion signal fitting residual value, the amplitude change trend parameter and the fitting regression level information, and is embedded in the output signal stream in real time and transmitted to the fault response unit; The delayed behavior status label is automatically updated every two sampling cycles to assist the fault response unit in executing the response level determination process. When the fitting regression level in the status label shows a deteriorating trend or the residual value continues to expand, the fault response unit loads the response judgment condition in advance and enters the candidate preparation state; The delay behavior status label is also used as one of the input fields for generating the fault event identification label. Its content is written into the event identification structure after the fault response behavior is triggered to support the subsequent external fault trend analysis and multi-label clustering process. Through the generation and dynamic update of the delay behavior status label, a data sharing mechanism is established between the protection delay unit and the fault response unit to realize the feedforward judgment of the response path and cross-unit linkage.

[0075] In a preferred embodiment of the present invention, after the protection delay unit enters the delayed state, in order to support the system to track the current delay behavior and optimize the decision, a structured data description information, called a delay behavior state label, is generated. The label is used as a dynamically updated state identification carrier to provide a quantitative description of the current state to other functional units of the system during the protection delay stage, and as an auxiliary basis for subsequent fault judgment and response triggering.

[0076] The "delay behavior status label" refers to a data set containing key judgment parameters generated during the delay response phase. The label mainly includes the following four fields: Delay time interval: refers to the cumulative time that the current delay state has been running since it was started, as well as the maximum allowable delay time range set by the system, in milliseconds; Current fusion signal fitting residual value: the current error value between the fusion signal and the regression prediction trend, which is used to reflect whether the current signal state is stable; Amplitude change trend parameter: reflects the change direction and rate of the current fusion signal in a continuous cycle, such as whether it is in an increasing trend or whether there is an accelerated rise phenomenon; Fitting regression level information: The regression trend level given by the error fitting unit after evaluation can usually be divided into three levels: "stable", "weak change" and "deterioration", corresponding to different degrees of response recommendations.

[0077] These parameter fields together constitute a complete tag information. Each tag is time-sensitive and only represents the system status within the current two sampling periods.

[0078] Real-time embedding and cross-unit transmission mechanism of tags: The tag is automatically refreshed every two sampling cycles, that is, the field value is re-evaluated and updated every two minimum detection cycles to ensure that the system has the "latest state snapshot". The tag data is embedded in the output signal stream through the internal communication link, that is, the tag information is attached to the data output channel of the main system for transmission to the fault response unit.

[0079] Through this mechanism, the system realizes data linkage between the protection delay unit and the fault response unit, so that the response behavior not only depends on the final judgment conclusion, but also has the ability to perceive the "current delay state evolution trend" and has the feedforward control capability of the response path.

[0080] The auxiliary role of the tag for the fault response unit: After receiving the tag, the fault response unit extracts its field content to assist in the response level determination process. The response level refers to the intensity of the response behavior selected by the system based on the judgment of the fault degree, which commonly includes signal bypass, data freeze, forced disconnection, etc.

[0081] The following fields in the tag have direct triggering significance: when the fitting regression level shows a deteriorating trend, that is, the system has identified an increase in the trend slope for consecutive periods, indicating that the fault is getting worse; or when the residual value continues to expand, that is, the fusion signal deviates further and further from the reference model, indicating that the error evolution is risky; when any of the above conditions is met, the fault response unit no longer waits for the delay state to end, but loads the response judgment condition in advance, that is, the current state is upgraded to the "candidate response state" and is ready to enter the formal response process. This mechanism improves the system's response agility to critical states, so that the system response behavior is no longer "passively waiting", but has the ability to reason and judge in advance.

[0082] The role of labels in the fault event identification structure: The delay behavior state label is not only a reference for judgment, but also a component field of the fault event identification label generation process. When the system finally enters the fault response process, the generated event identification structure will contain complete response background information, including: the current trigger action type; system runtime reference model information; fitting trend evolution path; and historical content records of the delay behavior state label. This information is output as a data encapsulation of a single event to support subsequent system-level fault trend analysis and label clustering comparison, and is used to identify high-frequency fault sources, optimize prediction models, and other purposes.

[0083] For example: During a backup power supply switching process, the system detected a medium-amplitude critical fluctuation in the current signal, and the protection delay unit entered the delayed state, with the initial delay time set to 15 milliseconds. After the label is generated, the content is: delay time interval: 0–15ms; current residual: 0.045; amplitude trend: increasing; fitting level: weak change. Updated after two cycles: delay time interval: 0–17ms; current residual: 0.061; fitting level changed to: deterioration. At this time, the fault response unit loads the response judgment condition, evaluates it as a high risk in advance, and marks it as a candidate response state. Subsequently, at the 18th millisecond, the system jumps to the response stage, executes signal bypass, and generates an event label, including the above-mentioned delay label field, for background system clustering analysis.

[0084] The fault response unit introduces the delayed behavior state label as the preprocessing information source for response judgment before executing the emergency response process. The fitting regression level and residual trend in the state label are used as one of the judgment factors. The offset degree, duration and fitting trend change between the current fusion signal and the reference model are combined to jointly determine the response behavior level. The response behavior is divided into three categories: interrupt output, signal bypass and data freeze. If the residual value in the delay behavior state label increases continuously and reaches five cycles, and the fusion signal deviation degree is greater than 15% for three consecutive cycles, the interrupt output is triggered; If the tag trend does not deteriorate significantly but the fluctuation lasts less than ten milliseconds, the data freeze strategy is executed; If the fitting trend shows nonlinear growth and the status label level is medium risk, the signal bypass logic is executed to redirect the current data flow to the redundant path output channel; this response mechanism introduces delayed behavior status labels as linkage conditions to improve the stability and reliability of response trigger judgment.

[0085] In the implementation of the present invention, the fault response unit is the core structure responsible for determining the final protection strategy execution mode when an abnormal event occurs during the current transformer monitoring process. In order to improve the judgment accuracy and avoid erroneous protection response caused by a single data error, the fault response unit first introduces the delay behavior state tag as a pre-processing information source for response judgment before executing the emergency response process.

[0086] The delay behavior status label is a structured status data packet generated by the protection delay unit when it is in the response delay stage, which contains multiple fields that reflect the current signal status and trend characteristics. Among them, the key fields particularly used for fault response pre-judgment include: Fitting regression level: indicating the level classification of the current error fitting trend, generally set to "stable", "medium risk" and "deterioration" to reflect whether the system trend is accelerating; residual trend: indicating whether the error between the fusion signal and the reference model continues to grow in the current few cycles. This information provides the fault response unit with external input to determine whether the system is "in the process of deterioration", and forms a multi-source joint judgment basis in conjunction with the signal data collected by itself.

[0087] The "fusion signal" is a high-reliability output data stream constructed by combining the main path sampling value and the backup path high-speed sampling value with time synchronization, phase calibration, and distortion elimination in the error fitting unit, and is used to reflect the current status in real time.

[0088] When judging whether to execute a response behavior, the system not only refers to the delayed behavior state label, but also comprehensively analyzes the following: the degree of offset between the fusion signal and the reference model: that is, the percentage change of the current signal relative to the normal value in multiple cycles; the duration of the offset: whether the offset state is maintained for a certain period to avoid triggering protection due to transient anomalies; the change of the fitting trend: refers to the rate of change of the current error trend, whether it is linear growth, nonlinear fluctuation or stabilization. These factors and the label field together constitute a multi-factor decision system to improve the stability and confidence of the response behavior judgment.

[0089] Response behavior level classification and execution conditions: In the present invention, response behaviors are subdivided into the following three types: Definition of interrupt output: immediately interrupt the current signal output, usually corresponding to the strongest protection level; trigger condition: the residual value in the delayed behavior state label increases continuously and reaches five cycles; at the same time, the offset degree of the fused signal is greater than 15% in three consecutive cycles; significance: it is judged that the fault trend is clear and aggravated, and the system safety must be protected immediately.

[0090] Data freeze definition: suspend data refresh, maintain current output value, and prevent further intervention; trigger condition: no obvious deterioration of label trend (fitting level is stable or medium risk); at the same time, the fluctuation duration is less than ten milliseconds, which is judged as a short-term disturbance; significance: judged as short-term jitter or pseudo-fluctuation, no need to perform disconnection or bypass, only freeze for observation.

[0091] Signal bypass definition: switch the output path, turn off the main path output signal, and use the backup path data stream as the main output; trigger condition: the fitting trend shows nonlinear growth, that is, the error slope is exponential and curve-shaped acceleration; at the same time, the state label level is medium risk; significance: it is considered that there is an abnormality in the main path, and the backup path data is still credible, and the path switching is used to avoid affecting the overall control strategy. Through the above classification, the system realizes a step-by-step protection system from low intensity to high intensity, ensuring the provision of appropriate response strategies in a variety of complex situations.

[0092] Assume that the system identifies a fusion signal fluctuation during a power switching process, and the protection delay unit enters the delayed state and generates a status label: The current cycle label shows: Fitting level = medium risk, residual trend = growth; After entering the fifth cycle, the residual growth continues; The fusion signal offset amplitude is 18%; The system compares the above conditions and meets the interrupt output trigger condition, immediately enters the emergency response process, executes the output interrupt, and cuts off the main output. In another scenario, the residual value in the label is not large, the fitting level is stable, and the duration is only 6ms, then the system determines it as a slight disturbance, only performs data freezing, and does not perform strong protection operations.

[0093] The fault response unit generates a fault event identification tag when triggering the protection behavior. The tag structure includes fault type code, fluctuation behavior tag, protection action record, fusion signal fitting trajectory information and delay behavior status tag field. It is encapsulated and output synchronously in the standard communication data frame format. Each event identification tag has a unique number for fault tracing and historical archiving. When there are similar fitting trends, similar delay behavior status fields and different fluctuation starting points in multiple event identification tags, the fault trend clustering logic is triggered, and an early warning report is sent to the superior platform.

[0094] In the implementation of the present invention, in order to achieve structured recording, response tracing and historical archiving of fault behaviors, the fault response unit generates a data structure with complete information and standardized structure when triggering the protection behavior, which is called the fault event identification tag. The tag is constructed while the protection action is executed, and is encapsulated and synchronously output in a standard communication data frame format for sharing between systems, platform recording and upper-level platform monitoring.

[0095] In a preferred embodiment of the present invention, the offset judgment standard between the fusion signal and the reference data model is calculated by percentage relative offset. The system sets 10% as the warning zone boundary and 15% as the fault response limit threshold. That is, when the offset value of the fusion signal exceeds 15% of its reference model value in three consecutive cycles, the system considers that the offset has exceeded the preset limit and triggers the fault response unit to generate a high-level protection instruction.

[0096] The fault event identification tag contains the following key fields, each of which has a clear technical function: Fault type coding: The system assigns a classification code to each event based on the fault behavior category; Function: Quickly identify types such as current overload, phase misalignment, sudden disturbance, and logic jump; Use: Used for classification query, statistical reports, and label clustering basic indexes.

[0097] Fluctuation behavior label: reflects the behavioral characteristics of the fusion signal in the identification stage before the event; includes: fluctuation amplitude range, duration classification, slope level; function: supports the judgment of whether the fault is triggered by the accumulation of continuous slight fluctuations or caused by a one-time drastic jump.

[0098] Protection action record: record of the response type and action level ultimately executed by the system; scope includes: interrupt output, signal bypass, data freeze; record content: response level, execution time, action duration.

[0099] Fusion signal fitting trajectory information: residual trend curve and regression model data summary output by the error fitting unit; usually adopts a simplified structure, such as: slope value sequence, regression level transfer sequence, residual extreme value, etc.; Function: Provide quantitative indicators for fault trend identification, support behavior analysis and model training.

[0100] Delay behavior status label field: protects the complete content of the label generated by the delay unit in the delayed response phase; includes: delay time, status level, trend evaluation value; function: reveals whether the system experienced a waiting period before the event and whether there is latent evolution in the trend.

[0101] At the same time when the protection action occurs, all the above field information is encapsulated by the fault response unit into a standard communication data frame format, which usually includes a frame header, a tag body field, a check bit and a frame tail, to ensure that the tag has consistent identification and compatibility in the system bus or network. The tag is synchronously output to: the upper platform (such as a data center centralized control system); the local event recording module; the fault history archive database. Each event identification tag has a unique number, and the number generation rule can be based on timestamp + fault type bit code + system node number to achieve full life cycle fault tracing capabilities.

[0102] Judgment conditions and action mechanism of fault trend clustering logic: The system accumulates a large number of fault event identification tags during continuous operation. In order to improve the fault early warning capability, the present invention designs a set of fault trend clustering logic to identify potential "repetitive but hidden" fault types in the system.

[0103] Clustering is triggered when the system detects the following three similarity features in multiple event identification tags: Similar fitting trends: There is trend slope and residual path consistency between the fusion signal and the fitting trajectory information in the tag; Similar delay behavior state fields: The delay tag content in the tag (such as delay interval, level mark, trend direction) is statistically clustered; Different fluctuation starting points: Multiple tags are generated at different time points, different load conditions or power supply states, indicating non-accidental consistency.

[0104] Once the clustering trigger conditions are met, the system will consider that there is a potential systemic risk evolution pattern, and then: send an early warning report to the superior platform (such as the data center monitoring and management system); the report includes: cluster label number, behavior feature summary, historical distribution curve, recommended response strategy, etc.; support operation and maintenance personnel to conduct preventive inspections, model revisions or system structure optimization.

[0105] Example description: Within three days, the system recorded 7 event identification tags, of which 5 tags met the following characteristics: the slope of the fitted trend was positively increasing; the residual value continued to rise during the delay phase; the status level in the delay tag was "medium risk"; and it occurred in different time periods and during peak load periods of different servers. The system determines that the cluster has consistent characteristics, triggers the clustering logic, and generates an early warning report to the superior platform, indicating that there may be uneven current distribution between systems or abnormal power switching strategy, and recommends investigation.

[0106] It should be noted that: in order to facilitate the understanding of the structural composition and response process of the current transformer intelligent control system proposed in the present invention, some control terms and system behavior instructions involved in the specification are defined and technically explained to ensure the consistency of the system structure and the collaborative logic between the functional units is open and clear.

[0107] "High-speed sampling instruction" refers to the trigger instruction issued by the fluctuation identification unit to the redundant acquisition unit after the system identifies that the current state is in a critical fluctuation type. It is used to start the backup sampling path and increase the sampling frequency to a higher rate than the set sampling rate of the main path, so as to achieve the capture of details of the current state during the fluctuation period and redundant verification. The triggering basis of this instruction is that the fluctuation identification unit's judgment result on the fusion signal characteristics meets the following conditions: the fluctuation intensity is within the system tolerance threshold, the frequency has not seriously shifted, and the trend has not shown continuous acceleration deviation. Although this type of fluctuation is not an immediate failure, it has the possibility of disturbance, so it is necessary to start the backup sampling to obtain higher-precision data. The high-speed sampling instruction has a one-time, single-cycle triggering feature. The sampling behavior is released after being completed on demand by the sampling drive logic in the redundant acquisition unit, and does not constitute continuous occupation of resources.

[0108] "High-priority protection response instructions" refer to a type of potential control instructions that can be called but not executed during the protection delay period, which are preset according to the error fitting unit and the delay judgment mechanism when the system identifies the risk of possible failure. This instruction is essentially a protection action response capability, such as current interruption, output cut-off, trip signal sending, etc., but when the fusion signal state has not completely deteriorated and the fitting trend has not formed a clear abnormality, it will be locked and delayed by the behavior priority conversion structure in the protection delay unit. The behavioral characteristics of this instruction are: "It has the execution preparation conditions but is suppressed by the system strategy logic" until the delay state is released or the response conditions are upgraded, and the fault response unit takes over and determines whether it is converted into a formal execution instruction (i.e., a high-level protection instruction). Through the preset and suppression mechanism of high-priority instructions, the system establishes a response "waiting area" for protection actions, effectively preventing false operations when data judgment is not sufficient, and improving the response accuracy of the system.

[0109] "High-level protection instructions" are execution-type control commands that are ultimately generated and sent by the fault response unit when the system determines that the current current state has reached an intolerable level during the fault response phase. Its corresponding response behaviors include but are not limited to: signal interruption output (termination of the current control output signal flow); signal bypass (switching to the backup path output); data freeze (maintaining the current output and suspending updates), etc. The triggering conditions of this instruction are usually that multiple independent dimensions reach the execution threshold, including but not limited to: the residual value in the delay behavior state label continues to grow and exceeds the set cycle threshold; the fitting trend shows nonlinear accelerated changes; the offset amplitude between the fusion signal and the reference model is greater than the set tolerance. Once the high-level protection instruction is executed, it usually has irreversibility, system mandatory execution and cross-module synchronous response characteristics. The response result is used as part of the fault event record content, embedded in the event identification label output, and transmitted to the upper-level platform or storage system to form a system protection behavior closed loop.

[0110] To ensure that the definition of the above control signals does not exist in isolation, the system establishes the following call sequence logic in the structural design: After the fluctuation identification unit determines the critical fluctuation → triggers the high-speed sampling instruction → the redundant acquisition unit samples the fusion signal; the fusion signal is analyzed by the error fitting unit → if the trend does not deteriorate → a high-priority protection response instruction is generated, but it is locked by the delay mechanism; if the trend escalates or the delay fails → the fault response unit generates a high-level protection instruction and enforces the corresponding response behavior. This structure ensures that the system has a complete closed loop from identification → suppression → confirmation → execution on the response path. The three types of control signals each have a trigger source, execution target and system interface position, forming a control model with clear functions, mutually exclusive strategies and non-conflicting behaviors.

[0111] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0112] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean 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 the present application.

[0113] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example 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 performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0114] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0115] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A current transformer intelligent control system, characterized in that: include: The fluctuation identification unit is used to obtain the current signal during the switching period through real-time sampling, extract the continuous change amplitude, rising edge slope, phase difference change and frequency mutation value from the current signal based on the predefined instantaneous change identification standard, and complete the feature identification and classification of abnormal fluctuation behavior according to data statistical rules; The redundant acquisition unit, after the identification result belongs to the tolerable switching fluctuation type, immediately triggers the backup sampling path and enables the high-speed sampling instruction to obtain the redundant current information flow, performs time axis comparison on the main path data, and constructs the fusion signal result on the premise of satisfying the timing continuity, which is used to replace the distortion or drift segment in the main signal; The error fitting unit continuously compares the fusion signal with the preset reference data model, performs fitting regression on the error deviation trend of the current fluctuation state, sets the protection trigger threshold and response time according to the fitting result, and makes real-time corrections to the main control strategy; The protection delay unit automatically delays the protection execution time when it determines that the fluctuation state is within the acceptable range, and outputs a fusion signal as an effective monitoring benchmark. At the same time, it suspends high-priority protection response instructions to prevent false operations caused by transient errors and improve the power supply continuity during switching. The fault response unit immediately executes the emergency response process, generates a high-level protection instruction and interrupts the signal output process when the offset between the fusion signal and the reference data model exceeds the preset limit, or the redundant sampling path cannot complete the effective time alignment, and sends a fault event identification to the external system for protection processing and log recording of power switching control.

2. A current transformer intelligent control system according to claim 1, characterized in that: The fluctuation identification unit includes an abnormal current fluctuation screening structure, which is used to set a multi-threshold condition judgment mechanism based on the extraction of continuous change amplitude, rising edge slope, phase difference change and frequency mutation value. Different characteristic parameters are combined and operated in a time-aligned manner to construct a high-dimensional characteristic behavior discrimination space. Through the set three-value hierarchical decision logic, the detection results are marked as stable fluctuation, critical fluctuation or severe abnormality. In the time window slip judgment process, whether the fitting coefficient between the average amplitude change and the rising rate of three adjacent sampling periods is lower than 0.1 is used as the first judgment factor, and whether the instantaneous frequency change is within 5 Hz is used as the second judgment factor. When both conditions are met at the same time, the current state is judged to be a critical fluctuation. The critical fluctuation type is identified as a tolerable switching fluctuation type, which is used as the basis for starting the subsequent backup sampling path, so that the working behavior of the redundant acquisition unit has a traceable judgment basis based on the feature recognition results, thereby ensuring the continuity of the technical chain between fluctuation judgment and data compensation.

3. A current transformer intelligent control system according to claim 2, characterized in that: The redundant acquisition unit builds a data mirror relationship by running the dual-path current signal channel in parallel, where the setting error between the main path sampling period and the backup path sampling period does not exceed ten microseconds to ensure that the time alignment accuracy meets the data fusion requirements; The backup path sampling values ​​need to go through a preliminary noise elimination process, where the elimination method includes one of the three methods: maximum difference value method, root mean square limit method, and adaptive judgment based on sliding window filtering; Under the condition that the sampling values ​​satisfy the redundant timing structure, an interleaved data stream is established by delaying one cycle to ensure the continuity of the redundant data and to ensure that it does not cause logical interference with the main path. At the same time, all redundant data will be brought into the phase reconstruction operation process to participate in the construction of the error fitting unit behavior.

4. A current transformer intelligent control system according to claim 3, characterized in that: The error fitting unit contains a dynamic regression correction structure, in which the regression behavior is analyzed based on the mean deviation trend of the fusion signal from the reference model in the current five sampling periods, while considering the consistency of the deviation direction and the stability of the duration; The regression strategy adopts one of two methods: interval minimum variance fitting method and exponential weighted moving average method. The former is used when the current fluctuation intensity is greater than ten percent, and the latter is used when the fluctuation duration is less than ten milliseconds. The offset is converted into a trigger adjustment amplitude factor, whose value range is 0.05 to 0.15, and is used to correct the protection trigger threshold.

5. A current transformer intelligent control system according to claim 4, characterized in that: The error fitting unit also includes a trend segment tracking structure, which is used to separate the influence of high-frequency disturbances. In each detection cycle, the behavior trend is analyzed by fitting residuals. When the residual is less than the reference value of 0.03 for three consecutive cycles, the fitting trend conservative correction stage is entered. In this stage, no rapid threshold regression adjustment is performed, and the protection strategy is updated only based on the historical cumulative average error. When the residual reverses direction twice within five cycles, the automatic output signal does not participate in the threshold correction and only maintains the current delay behavior.

6. A current transformer intelligent control system according to claim 5, characterized in that: The protection delay unit includes a behavior priority conversion structure. When the fusion signal is classified as a critical fluctuation type, the protection execution time interval is actively delayed. The delay value is not less than five milliseconds and not more than twenty milliseconds. The specific delay time is obtained by linear mapping based on the adjustment amplitude factor output by the error fitting unit. During the delay period, all signal outputs are marked as medium risk, and high-level interrupt control commands are prohibited from being sent; after the delay, if the regression value of the error fitting unit remains stable, the delay will continue; otherwise, the protection action call behavior in the fault response unit will be executed.

7. The intelligent control system for current transformer according to claim 6, characterized in that: After entering the delayed state, the protection delay unit generates a delay behavior state label, which contains the delay time interval, the current fusion signal fitting residual value, the amplitude change trend parameter and the fitting regression level information, and is embedded in the output signal stream in real time and transmitted to the fault response unit; The delayed behavior status label is automatically updated every two sampling cycles to assist the fault response unit in executing the response level determination process. When the fitting regression level in the status label shows a deteriorating trend or the residual value continues to expand, the fault response unit loads the response judgment condition in advance and enters the candidate preparation state; The delay behavior status tag is also used as one of the input fields for generating the fault event identification tag, and its content is written into the event identification structure after the fault response behavior is triggered.

8. The intelligent control system for current transformer according to claim 7, characterized in that: The fault response unit introduces the delayed behavior state label as the preprocessing information source for response judgment before executing the emergency response process. The fitting regression level and residual trend in the state label are used as one of the judgment factors. The offset degree, duration and fitting trend change between the current fusion signal and the reference model are combined to jointly determine the response behavior level. The response behavior is divided into three categories: interrupt output, signal bypass and data freeze. If the residual value in the delay behavior state label increases continuously and reaches five cycles, and the fusion signal deviation degree is greater than 15% for three consecutive cycles, the interrupt output is triggered; If the tag trend does not deteriorate significantly but the fluctuation lasts less than ten milliseconds, the data freezing strategy is executed; If the fitting trend shows nonlinear growth and the status label level is medium risk, the signal bypass logic is executed to redirect the current data flow to the redundant path output channel.

9. The intelligent control system for current transformer according to claim 8, characterized in that: The fault response unit generates a fault event identification tag when triggering the protection behavior. The tag structure includes fault type code, fluctuation behavior tag, protection action record, fusion signal fitting trajectory information and delay behavior state tag field. It is encapsulated and output synchronously in the standard communication data frame format. Each event identification tag has a unique number for fault tracing and historical archiving. When there are similar fitting trends, similar delay behavior status fields, and different fluctuation starting points in multiple event identification tags, the fault trend clustering logic is triggered and an early warning report is sent to the superior platform.

Images