Self-calibration and drift compensation method for flow velocity baffle of transformer gas relay

Abstract

The invention relates to a self-calibration and drift compensation method for a flow velocity baffle of a transformer gas relay. The method comprises the following steps of: acquiring an alarm state, a tripping state and a cavity pressure change of a gas relay as well as a short-circuit current event, tripping sequence information and oil temperature transition information of a transformer; when the pressure of the cavity continuously rises in one direction within the preset time, the gas accumulation risk is judged, when the flow velocity baffle continuously deflects within the preset time and the contact is increased in a critical jitter mode, the strong oil flow risk is judged, and self-calibration is forbidden if any risk is true; when the gas accumulation risk and the strong oil flow risk do not exist, entering a self-calibration session and locking a flow speed baffle action threshold value setting parameter; during the session, only read-only sampling is carried out, and the calculated threshold compensation amount is written into a shadow parameter area; self-calibration is prohibited actively when a gas accumulation risk or a strong oil flow risk exists, so that the action threshold is prevented from being adjusted mistakenly under unstable working conditions such as a real fault, external electric disturbance or strong oil flow impact.

Description

Technical Field

[0001] This invention relates to a self-calibration and drift compensation method, specifically a self-calibration and drift compensation method for a flow rate baffle in a transformer gas relay. Background Technology

[0002] Based on the publicly available technical solutions, the online testing methods for gas relays, represented by CN112630637B and CN107478983A, have improved the convenience of testing under uninterrupted power conditions to some extent. However, if they are used for the self-calibration and long-term drift compensation of flow rate baffles, there are still obvious shortcomings and engineering drawbacks.

[0003] First, the core of these solutions remains online verification rather than operational self-calibration. Their technical approach primarily relies on external variable frequency oil pumps, flow meters, oil reservoirs, T-junctions, and multiple sets of solenoid valves. They artificially construct oil flow or pressure conditions to verify whether the gas relay operates or is up to standard. This method is essentially an active excitation-based detection process, rather than adaptive parameter correction based on natural operating data under real-world conditions. Therefore, it cannot solve the problem of slow threshold drift caused by spring fatigue, mechanical wear, and temperature changes during long-term service of the flow velocity baffle. Second, the aforementioned comparative documents generally require structural modifications to the transformer's original oil circuit, introducing additional oil pumps, valves, and inspection branches. On main transformers with voltage levels of 220kV and above, such modifications not only have long construction cycles and complex approval processes but also increase potential risks of oil leakage, valve misoperation, and oil pump failure. From an operation and maintenance perspective, this is not conducive to large-scale promotion and is even less suitable as a long-term, routine online drift compensation method. Secondly, the oil flow or surge conditions simulated by these schemes are relatively idealized and differ significantly from the complex dynamic processes in actual power grid operation, which are formed by the superposition of factors such as short-circuit current impacts, tap changer actions, load changes, trip sequences, and rapid oil temperature jumps. In particular, phenomena commonly seen in real operation, such as transient pressure changes, short-term reverse oil flow, and critical contact jitter, are often difficult to fully reproduce under the simulated conditions. This leads to a discrepancy between the verification conclusions and the actual false alarm behavior, easily resulting in situations where verification is passed but false alarms still occur during operation. At the same time, existing comparative schemes lack systematic constraints on self-calibration safety boundaries. Their processes do not establish clear risk gating logic, do not strictly distinguish between gas accumulation risks and strong oil flow risks, and do not combine short-circuit current events and trip sequences for counter-evidence judgment and disturbance elimination. When pressure or oil flow anomalies originate from external electrical disturbances, it is difficult to effectively eliminate them at the algorithm level. This can easily lead to misjudgments in actual operation, thus making threshold adjustments based on unsafe or unrealistic data.

[0004] Furthermore, the outputs of the aforementioned solutions largely remain at the level of judging whether the action is qualified, without forming a threshold compensation amount that can be directly used for updating tuning parameters. They also lack a conversational self-calibration process, shadow parameter area caching, and a mechanism for writing back after consistency verification. In practical applications, they still rely on manual experience for tuning and correction, which is not only inefficient but also difficult to unify adjustment standards among different personnel and sites, potentially introducing new uncertainties over long-term operation. In addition, existing technologies generally ignore the impact of oil temperature changes on the flow rate baffle action threshold and do not require consistency verification of compensation results under different oil temperature transition states. This can easily lead to over-correction within a single temperature range, amplifying the risk of false alarms or failure to operate under seasonal changes or extreme temperature differences. Regarding data reliability, the comparative solutions lack a quantitative evaluation system for sampling credibility, cannot classify and manage data quality based on the stability of risk assessment, and cannot distinguish which results are only suitable for observation and reference and which have actual write-back value. This results in threshold updates being either too conservative and ineffective in the long term, or too aggressive and prone to write errors. Summary of the Invention

[0005] The purpose of this invention is to provide a self-calibration and drift compensation method for a flow rate baffle in a transformer gas relay, thereby addressing some of the drawbacks and shortcomings mentioned in the background art.

[0006] The present invention adopts the following technical solution to solve the above-mentioned technical problems:

[0007] Collect gas relay alarm status, trip status and cavity pressure changes, as well as transformer short circuit current events, trip sequence information and oil temperature transition information;

[0008] When the cavity pressure continues to rise unidirectionally within a preset time, it is determined to be a risk of gas accumulation. When the flow rate baffle continues to deflect within a preset time and the contact point shows a critical increase in vibration, it is determined to be a risk of strong oil flow. If either risk is met, self-calibration is prohibited. When there is no risk of gas accumulation and no risk of strong oil flow, the self-calibration session is entered and the flow rate baffle action threshold setting parameter is locked. During the session, only read-only sampling is performed and the calculated threshold compensation amount is written to the shadow parameter area.

[0009] When the threshold compensation amount remains consistent across multiple sampling periods, the threshold compensation amount is written into the update action threshold tuning parameter; otherwise, the original tuning is maintained. The session time window and sampling reliability identifier are output for subsequent drift compensation constraints.

[0010] Furthermore, the determination of the risk of gas accumulation satisfies the monotonicity of pressure change and the counter-evidence condition within a preset time. The monotonicity is that the cavity pressure rises in the same direction for multiple consecutive samplings without falling back. The counter-evidence condition is that no transient disturbance corresponding to a short-circuit current event or trip sequence is detected within the preset time. The determination of the risk of strong oil flow also satisfies the dual evidence condition within the preset time. The dual evidence condition is that the continuous deflection of the flow velocity baffle and the increase in the critical vibration of the contact point occur simultaneously and are synchronized in time.

[0011] Furthermore, when the threshold compensation amount is written to the shadow parameter area during the session, a sampling trust identifier is generated and hierarchical writing control is implemented. When the sampling trust identifier reaches a preset level, a switchable version is formed in the shadow parameter area. When the preset level is not reached, only a non-switchable draft is retained. And when the risk of gas accumulation or strong oil flow is detected again during the self-calibration session, the switchable version is frozen and the draft is discarded.

[0012] Furthermore, the multiple sampling periods include at least two sampling periods under different oil temperature transition states, and the threshold compensation amount is required to change in the same direction and the difference does not exceed the consistency threshold. Only when this condition is met will the update action threshold tuning parameter be written. After the update is written, the session time window and the sampling trust identifier are stored, and the threshold compensation amount for the next write update is limited to not exceeding the maximum step amount determined by the sampling trust identifier.

[0013] Furthermore, the cavity pressure is continuously sampled within the preset time and short-term reverse fluctuations are suppressed. When an instantaneous drop occurs that does not exceed the preset duration and its amplitude is less than the preset drop threshold, it is still determined as no drop.

[0014] Furthermore, the counter-evidence condition includes a disturbance fingerprint exclusion logic. When a short-circuit current event or trip sequence information is detected, the cavity pressure change segment that is aligned with its time within the preset time period is marked as an external electrical disturbance segment and removed from the monotonicity judgment. The gas accumulation risk is only determined if the pressure continues to rise in the same direction after removal.

[0015] Furthermore, the sampling confidence identifier is generated by the risk discrimination stability within the self-calibration session. This risk discrimination stability includes ensuring that both the gas accumulation risk and the strong oil flow risk remain false and do not undergo state reversal within a preset duration. This stability is mapped to a confidence level. Only when the confidence level reaches a preset level is the switchable version allowed to be formed in the shadow parameter area. The confidence level is determined by a confidence score C, which is calculated as follows:

[0016]

[0017] in, This represents a sequence of risk states within a preset duration, and This indicates that either the risk of gas accumulation or the risk of strong oil flow is present. This indicates that neither of the two types of risks applies; For the risk state sequence The number of state transitions from 0 to 1 or from 1 to 0 within the preset duration is used to characterize the stability of risk assessment. , which is the flip penalty coefficient, used to adjust the inhibitory strength of state flipping on the credibility score C; The time window length corresponding to the preset duration; Within the time window, the following conditions must be met. The cumulative duration is used to characterize the stable persistence of the two types of risks, which does not hold true. This is a margin for risk of gas accumulation. Both are used to characterize the safety margin between the risk of strong oil flow and their respective judgment thresholds when the risk does not materialize, and The minimum safety margin between the two types of risks is used as a constraint for credibility calculation; when the credibility score C satisfies the preset threshold range and the credibility level reaches the preset level, the switchable version is allowed to be formed in the shadow parameter area; otherwise, only the non-switchable draft is allowed to be formed or the switchable version is prohibited from being formed.

[0018] Furthermore, the freeze operation when the risk of gas accumulation or the risk of strong oil flow is detected includes version sealing and write-back prohibition. Version sealing is to mark the most recently formed switchable version in the shadow parameter area as unwritable and record the corresponding session time window. Write-back prohibition is to prohibit the switchable version from being written to the update action threshold tuning parameter during the subsequent preset cooling period, until a sampling trust identifier that meets the preset level is regenerated.

[0019] Furthermore, the version sealing includes sealing reason binding and evidence snapshot recording. The sealing reason binding is to mark the risk type that triggers freezing as gas accumulation risk or strong oil flow risk and store it in association with the switchable version. The evidence snapshot recording is to record the cavity pressure change characteristics, flow baffle deflection characteristics and contact critical jitter characteristics at the moment of triggering freezing while recording the session time window.

[0020] Furthermore, the preset cooling period is determined using adaptive annealing logic. The adaptive annealing logic dynamically extends or shortens the cooling period based on the recovery process of the sampled trusted identifier. When the sampled trusted identifier does not reach the preset level within multiple consecutive sampling periods, the cooling period is extended. When the sampled trusted identifier reaches the preset level within multiple consecutive sampling periods and the risk judgment remains invalid, the cooling period is shortened.

[0021] The beneficial effects of this invention are as follows: By actively prohibiting self-calibration when there is a risk of gas accumulation or strong oil flow, it avoids erroneous adjustment of the action threshold under non-steady-state conditions such as actual faults, external electrical disturbances, or strong oil flow impacts, thereby preventing the threshold from being incorrectly learned and significantly improving the operational safety and engineering reliability of online self-calibration. In the determination of gas accumulation risk, this invention not only requires the cavity pressure to rise continuously in the same direction without falling back within a preset time, but also introduces a counter-evidence condition to explicitly exclude the influence of transient disturbances that are time-aligned with short-circuit current events or trip sequence. At the same time, by suppressing short-term, low-amplitude reverse fluctuations, the pressure determination has stronger robustness to non-essential factors such as sampling noise and transient oil temperature disturbances.

[0022] This effectively avoids misinterpreting external electrical disturbances or measurement noise as gas accumulation, reducing the probability of false alarms and false self-calibration inhibition by the gas relay. By simultaneously monitoring the continuous deflection of the flow velocity baffle and the increase in critical contact jitter, requiring both to occur synchronously, it avoids triggering strong oil flow risk judgments based solely on a single mechanical or electrical quantity. This dual-evidence mechanism can effectively distinguish between normal oil flow fluctuations and truly potentially jeopardizing the reliability of the gas relay's operation, reducing false judgments while ensuring timely inhibition of self-calibration when a genuine strong oil flow occurs, thus balancing sensitivity and selectivity. Attached Figure Description

[0023] Figure 1 The logic diagram for the self-calibration admission and threshold update of the gas relay is shown.

[0024] Figure 2 A diagram showing the relationship between multi-condition risk assessment and shadow parameter hierarchical writing functions.

[0025] Figure 3 Diagram showing the relationship between sampling trust identifier and freeze-cool writeback control function.

[0026] Figure 4 This is a schematic diagram illustrating the transition between cavity pressure and oil temperature, as well as the removal of disturbed fingerprints.

[0027] Figure 5 Plan view for determining the risk of strong oil flow using dual evidence.

[0028] Figure 6 This is a schematic diagram of the results of multi-time period consistency verification and threshold write-back.

[0029] Figure 7 This diagram illustrates the calculation of the credibility score C and its mapping to the credibility level.

[0030] Figure 8 The image shows the synchronized display of three features: freeze trigger and evidence snapshot.

[0031] Figure 9 The adaptive annealing cooling time variation and write-back unlock timeline. Detailed Implementation

[0032] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0033] Combined with appendix Figure 1 As shown, in this embodiment, the basic operating information for self-calibration and drift compensation is obtained by continuously monitoring the transformer gas relay and its associated operating status. First, the operating status of the gas relay itself is collected, including whether it is in an alarm state or a tripped state, to determine whether the current protection action has been initiated or is in a critical stage of imminent initiation. Simultaneously, the pressure changes within the gas relay cavity are collected in real time, and the pressure change trend over time is used to reflect the accumulation of gas or oil flow disturbances inside the relay.

[0034] While collecting the operating status information of the gas relay, the electrical and thermal operating information of the transformer is also collected simultaneously. Specifically, the system collects information on whether a short-circuit current event has occurred in the transformer to identify the instantaneous impact of external electrical faults on the pressure in the gas relay cavity and the operation of the flow velocity damper. Furthermore, the system collects the transformer's trip sequence information, and by analyzing the timing relationship and duration characteristics of the trips, it determines whether there are any abnormal operating processes related to relay protection operations. In addition, the system continuously monitors changes in transformer oil temperature, acquiring information on whether temperature jumps occur, as well as the magnitude and duration of these jumps, to reflect changes in the transformer's internal thermal state and their impact on the oil flow state.

[0035] The information is subjected to time alignment and continuity analysis to determine whether there are operational risks that would preclude self-calibration. First, pressure changes within the gas relay cavity are continuously monitored. When a continuous upward trend in the cavity pressure is detected within a preset time range without a significant drop, it is determined that gas is gradually accumulating within the cavity, thus identifying a risk of gas accumulation. During the determination process, the continuity and directionality of pressure changes are confirmed to avoid misinterpreting instantaneous fluctuations or sampling noise as gas accumulation.

[0036] Simultaneously, the motion state and contact behavior of the flow velocity baffle are jointly analyzed. When continuous deflection of the flow velocity baffle is detected within a preset time range, and an increased vibration amplitude occurs simultaneously with this deflection process, indicating that the contact is in a near-engaged or released state, it is determined that this operating state corresponds to a strong oil flow acting on the flow velocity baffle, thus identifying a strong oil flow risk. By simultaneously judging the mechanical deflection characteristics of the baffle and the electrical critical behavior of the contact, normal micro-disturbances and abnormal oil flow impacts can be effectively distinguished. When either the risk of gas accumulation or the risk of strong oil flow is detected, it is determined that the current operating state does not meet the self-calibration conditions, and the self-calibration process is immediately prohibited to prevent faults or abnormal operating conditions from being mistakenly used as calibration basis. When neither the risk of gas accumulation nor the risk of strong oil flow is detected within the preset time range, the current operating state is determined to be a non-fault stable state, allowing entry into the self-calibration session, and locking the action threshold setting parameter of the flow velocity baffle to prevent the parameter from being directly modified during the self-calibration session.

[0037] The effectiveness and stability of the threshold compensation amount are further verified. Specifically, the self-calibration session is divided into multiple independent sampling periods. Within each sampling period, the threshold compensation amount is calculated based on the corresponding operational data, and the threshold compensation amounts obtained from each sampling period are compared and analyzed. When the threshold compensation amounts obtained within the multiple sampling periods maintain consistency in the direction and amplitude range of change, and no significant dispersion or reverse change occurs, the threshold compensation amount is determined to be stable and repeatable.

[0038] After confirming that the threshold compensation amount meets the stability requirements, the write restriction on the flow velocity baffle action threshold tuning parameter is lifted, the threshold compensation amount is written, and the action threshold tuning parameter is updated so that the updated action threshold can reflect the true action characteristics of the flow velocity baffle under the current operating conditions. If the threshold compensation amounts obtained in the multiple sampling periods are significantly inconsistent or fluctuate, the self-calibration result is determined to be unreliable, the original action threshold tuning parameter is kept unchanged, and the current self-calibration session is terminated.

[0039] After updating the threshold tuning parameters or determining not to update, the session time window information and sampling reliability identifier corresponding to this self-calibration session are output. The session time window identifies the running time range on which this self-calibration is based, and the sampling reliability identifier characterizes the reliability of the self-calibration result. The session time window and sampling reliability identifier are recorded and invoked as constraints for subsequent drift compensation processes to limit the triggering timing and compensation magnitude of drift compensation, thereby ensuring the protection reliability and stability of the flow rate baffle during long-term operation.

[0040] Combined with appendix Figure 2As shown, in this embodiment, to improve the accuracy and anti-interference capability of risk assessment, multiple constraints are introduced for the assessment of gas accumulation risk and strong oil flow risk. For gas accumulation risk, continuous sampling and trend analysis are performed on the pressure changes in the gas relay cavity within a preset time range. When the pressure values ​​corresponding to multiple continuous sampling points consistently rise in the same direction and no significant drop occurs within this time range, the cavity pressure change is determined to meet the monotonicity requirement. By judging the direction and continuity of pressure changes, short-term pulsations or measurement noise are avoided from being misidentified as gas accumulation. Based on meeting the monotonicity of pressure changes, a counter-evidence condition is introduced to further eliminate the influence of external disturbances. Specifically, within the preset time range, whether a short-circuit current event or a tripping sequence occurs in the transformer is detected synchronously. When such electrical events are detected, the cavity pressure change at the corresponding moment is considered to be caused by external electrical disturbances and is not used as a basis for gas accumulation assessment. Only when no transient disturbance corresponding to a short-circuit current event or trip sequence is detected within a preset time, and the cavity pressure continues to rise in the same direction, is it finally determined that there is a risk of gas accumulation, thereby improving the reliability of gas accumulation determination.

[0041] To determine the risk of strong oil flow, a joint analysis is performed on the mechanical motion state of the flow velocity baffle and the electrical behavior of the contacts within a preset time range. When continuous deflection of the flow velocity baffle is detected within this time range, and simultaneously with this deflection process, an increase in the critical vibration amplitude of the contacts, indicating they are in a near-engaged or unengaged state, is observed, the flow velocity baffle is considered to be subjected to a strong oil flow. By requiring the continuous deflection characteristic of the baffle and the critical vibration characteristic of the contacts to occur simultaneously and remain synchronized in time, misjudgments caused by occasional vibrations or single signal anomalies can be effectively distinguished. Only when both types of evidence corroborate each other is a strong oil flow risk determined, thus providing a reliable basis for the safe triggering of subsequent self-calibration sessions.

[0042] The threshold compensation amount is written to the shadow parameter area, and a sampled trust identifier is generated simultaneously to characterize the reliability of the threshold compensation amount. The sampled trust identifier is determined by a combination of risk discrimination stability and data continuity within the self-calibration session, reflecting the stability of the operating state during the threshold compensation amount formation process. When writing the threshold compensation amount to the shadow parameter area, hierarchical writing control is implemented based on the generated sampled trust identifier. When the sampled trust identifier reaches a preset level, a switchable version is allowed to be generated in the shadow parameter area, making the threshold compensation amount eligible to be written into the updated action threshold tuning parameters when subsequent conditions are met. When the sampled trust identifier does not reach the preset level, the threshold compensation amount is only retained as a non-switchable draft for subsequent comparison and reference, but is not allowed to participate in the update of the action threshold tuning parameters, thereby avoiding the impact of low-reliability results on the protection function.

[0043] If a risk of gas accumulation or strong oil flow is detected again during the self-calibration session, the current self-calibration logic is immediately terminated, and the contents of the shadow parameter area are frozen. The freezing process includes marking the existing switchable version as unwriteable and discarding draft data that has not yet met the switchable conditions, restoring the shadow parameter area to a safe state before the risk occurred.

[0044] The multiple sampling periods include at least two sampling periods under different oil temperature transition states. One sampling period corresponds to a state where the oil temperature is basically stable or changes slowly, while the other sampling period corresponds to a state where the oil temperature rises or falls significantly. By obtaining the threshold compensation amount under different thermal conditions, the adaptability of the compensation amount to oil temperature changes can be verified, avoiding the direct use of compensation results formed under only a single thermal state for action threshold updates.

[0045] After calculating data for multiple sampling periods, a consistency analysis is performed on the threshold compensation values ​​obtained from each sampling period. The analysis requires that the direction of change for each threshold compensation value remains consistent, and that their differences do not exceed a preset consistency threshold. When both conditions are met, the threshold compensation value is determined to have consistency and stability under different oil temperature transition states, thus allowing the threshold compensation value to be written into and the action threshold tuning parameters of the flow rate baffle to be updated. If the consistency condition is not met, the original action threshold tuning parameters are kept unchanged, and the current self-calibration session ends.

[0046] After updating the action threshold tuning parameters, the session time window corresponding to this self-calibration session and the sampling trust identifier are stored together to record the operating conditions and data reliability on which this threshold update is based. Simultaneously, when subsequent threshold updates are allowed, the update magnitude of the threshold compensation is limited based on the stored sampling trust identifier, restricting the threshold compensation amount written in the next update from exceeding the maximum step size determined by the sampling trust identifier.

[0047] Time-series analysis is performed on the continuously sampled pressure sequence. When a momentary drop occurs during pressure change, lasting no more than a preset duration, and the drop amplitude is less than a preset drop threshold, this momentary drop is considered to be caused by measurement noise or short-term disturbance, and is not taken as evidence of a reversal in the pressure trend. Therefore, in this case, it is still determined that there is no drop in the cavity pressure change. By suppressing short-term reverse fluctuations, the continuity and stability of the pressure change trend judgment can be effectively maintained.

[0048] After pressure fluctuation suppression, a counter-evidence condition is introduced to further eliminate the influence of external electrical events on cavity pressure changes. Specifically, when a short-circuit current event or trip sequence information is detected in the transformer, time-aligned analysis is performed on the cavity pressure changes within the preset time range. The cavity pressure abrupt change segment corresponding to the occurrence time of the short-circuit current event or trip sequence is marked as an external electrical disturbance segment. Pressure data marked as external electrical disturbance segments are discarded when performing pressure change monotonicity judgment and are not included in the gas accumulation trend analysis. After discarding the external electrical disturbance segments, the continuity and directionality of the remaining cavity pressure sampling data are re-judged. Only when the cavity pressure still meets the condition of continuous upward movement in the same direction and no effective decline after discarding the disturbance segment is the risk of gas accumulation finally determined.

[0049] Combined with appendix Figure 3 As shown, in this embodiment, a sampling trust identifier generation mechanism is introduced to quantitatively evaluate the reliability of the threshold compensation amount formed during the self-calibration session. The sampling trust identifier is generated by the risk discrimination stability within the self-calibration session and is used to reflect the overall stability of the operating state within the session's time window. The risk discrimination stability is based on the condition that both gas accumulation risk and strong oil flow risk remain invalid and do not undergo state reversal within a preset duration. Through comprehensive analysis of the risk state's characteristics changing over time, stability is mapped to a trust level. Only when the trust level reaches a preset level is a switchable version allowed to be formed in the shadow parameter area.

[0050] Specifically, within the preset duration corresponding to the self-calibration session, a risk state sequence is constructed. Among them, the value is taken when there is a risk of gas accumulation or strong oil flow at any given time. When neither of the two types of risks applies, the value is taken as follows: By performing time-series analysis on the risk state sequence, the number of times the risk state changes from zero to one or from one to zero within the preset duration is counted and denoted as . The number of state flips is used to characterize the stability of the risk assessment process; the more flips, the more unstable the operating state.

[0051] Based on this, a credibility score is defined:

[0052]

[0053] in, The flip penalty coefficient is used to adjust the inhibitory effect of the number of state flips on the credibility score. When the risk state flips frequently, the credibility score is significantly reduced through exponential decay. The time window length corresponding to the preset duration. To meet the requirements within this time window The cumulative duration is used to characterize the stability ratio where the risk of gas accumulation and the risk of strong oil flow do not hold.

[0054] At the same time, a risk margin factor is introduced to reflect the safety margin, in which This indicates the margin of risk for gas accumulation. This represents the risk margin for strong oil flow; both represent the distance between the current operating state and their respective risk assessment thresholds. By taking... As a constraint, the credibility score is limited to the most unfavorable side of the safety margin, thereby avoiding obtaining a high credibility rating when a single risk is close to the threshold.

[0055] After the credibility score is calculated, it is mapped to the corresponding credibility level according to a preset threshold range. When the credibility level reaches the preset level requirement, a switchable version is allowed to be formed in the shadow parameter area, making the corresponding threshold compensation amount eligible to be written into the threshold tuning parameters of the subsequent update action; when the credibility level does not reach the preset level, only a non-switchable draft is allowed to be formed or a switchable version is prohibited from being formed.

[0056] Version sealing refers to marking the most recently formed switchable version within the shadow parameter area that meets the switchable conditions as unwritable, while simultaneously recording the self-calibration session time window corresponding to that version. By sealing the switchable version, its content remains unchanged during subsequent processing, preventing it from being overwritten or modified while preserving its integrity as a historical reference. The recorded session time window is used to identify the runtime range on which the switchable version was based, providing a time baseline for subsequent analysis of risk recovery. After version sealing is completed, a write-back prohibition operation is executed.

[0057] During the subsequent preset cooling period, the system continuously prohibits the switchable version from writing to the action threshold tuning parameters of the update flow rate baffle, even if the switchable version met the write conditions before freezing, it is not allowed to participate in the threshold update. The preset cooling period is used to ensure that the risk state completely subsides and the operating state stabilizes again, thereby avoiding premature adoption of self-calibration results before the risk has been fully eliminated. Only when the preset cooling period ends, the self-calibration session is re-entered, and a sampling trust identifier that meets the preset level is generated, will the write-back prohibition be lifted and the switchable version's write eligibility be restored.

[0058] When a risk of gas accumulation or strong oil flow is detected and a freeze operation is triggered, the reason for the freeze is first marked. The reason for freeze binding refers to explicitly identifying the risk type that triggered the freeze as either a gas accumulation risk or a strong oil flow risk, and associating this risk type with the corresponding switchable version for storage, so that each frozen version has a unique and clear freeze reason identifier. This method allows for differentiation of the impact of different risk sources on the self-calibration results during subsequent operation, avoiding misinterpretation of freeze cancellation when the nature of the risk has not changed.

[0059] While completing the binding of the sealing cause, an evidence snapshot is recorded of the operational status at the moment the freeze is triggered. This evidence snapshot is recorded simultaneously with the self-calibration session time window, including the cavity pressure change characteristics, the flow rate baffle deflection characteristics, and the jitter characteristics of the contacts when they are in a critical state at the moment the freeze is triggered. By simultaneously recording these multi-source characteristics, a complete set of evidence reflecting the operational status at the instant the freeze is triggered is formed.

[0060] After the freeze operation is triggered, the system continuously monitors the changes in the sampling trust identifier in subsequent sampling periods, and uses the recovery process of the sampling trust identifier as the main basis for adjusting the cooling period. When the sampling trust identifier fails to reach the preset level in multiple consecutive sampling periods, it indicates that the operating state has not yet recovered to a stable trust level. At this time, the cooling period is extended to continue to suppress the writing of self-calibration results, avoiding premature lifting of the write-back prohibition under unstable conditions.

[0061] When the sampling confidence level consistently reaches a preset level across multiple consecutive sampling periods, and both the risk of gas accumulation and the risk of strong oil flow remain invalid within the corresponding sampling periods, it indicates that the operating state has returned to a stable state. At this point, by shortening the remaining length of the cooling period or ending the cooling period early, the system can more quickly enter a state where the self-calibration results can be re-evaluated and applied. Through the aforementioned adaptive annealing logic, the cooling period can be automatically adjusted according to the actual changes in the operating state, thereby improving the overall operating efficiency of the flow rate damper self-calibration and drift compensation method while ensuring the reliability of the gas relay protection.

[0062] Example 1:

[0063] The main transformer protection system of a 220kV substation. The field equipment is a 220kV oil-immersed on-load tap-changing transformer with a capacity of 240MVA, model SSZ11240000 / 220. The matching gas relay is an electromechanical structure with a flow velocity baffle and contact mechanism and an external digital acquisition module. The equipment has been in operation for 6 years. In the past 3 months, it has experienced two unexplained alarms without differential protection activation. The maintenance personnel found that the alarm probability increased when the oil temperature fluctuated significantly during the day. The preliminary judgment is that the false alarm is caused by the flow velocity baffle threshold drifting due to mechanical wear and temperature.

[0064] The system composition and signal input are as follows: Alarm status signal A and trip status signal T are acquired from the gas relay side; the output P of the chamber pressure sensor and the output Temp of the oil temperature sensor are acquired; simultaneously, short-circuit current event marker SC and trip sequence information TripSeq are acquired from the protection device and fault recording system. The mechanical quantity of the flow velocity baffle is the deflection angle Theta output by the angular displacement sensor; the critical contact jitter is obtained by combining the contact voltage jitter and the opening / closing count to obtain the jitter index J. All quantities are sampled at a 1-second interval and entered into the self-calibration discrimination module, and system log entries Log are recorded synchronously for traceability.

[0065] The parameter settings use engineering-achievable example values: the preset time window length T is 600s, the sampling period is 1s, and the short-term reverse fluctuation suppression threshold is set to the duration of the instantaneous fallback. The threshold for the drop amplitude is 3 seconds. The Pa value is 0.2 kPa, and the consistency threshold epsilon is 3%. The baffle deflection threshold in the strong oil flow risk criterion is... The critical contact jitter threshold is 8.0%, expressed as a normalized percentage based on the rated deflection. The value is 0.65, and the jitter index is a normalized value between 0 and 1. Pressure monotonicity is represented by a first-order difference, defined as...

[0066]

[0067] If the vast majority of ( And it is permissible for a duration not exceeding [a certain number of times]. And the amplitude is less than A short-term pullback is still considered as no pullback. For ease of engineering calculation and reproduction, this embodiment adds a net increase threshold to continuous upward movements in the same direction. The minimum pressure is set to 0.8 kPa to ensure consistent judgment of the net pressure change after removing disturbances within the window.

[0068] Step S1 is data acquisition and preprocessing. Within each 600s window, the system acquires A and T, P and Temp to form a pressure sequence and a first-order difference sequence, and simultaneously acquires SC and TripSeq with timestamps aligned, and acquires Theta and J to form dual-evidence input. Step S2 is risk assessment. Gas accumulation risk assessment requires that the pressure continuously rises in the same direction without falling back within the window, and meets the disproving condition, i.e., no transient disturbance corresponding to SC or TripSeq is detected within the window. To ensure the verifiability of the disproving condition, this embodiment introduces disturbance fingerprint exclusion logic. If SC or TripSeq is detected, the pressure abrupt change segment aligned with its time is marked as an external electrical disturbance segment and removed from the monotonicity assessment. Only if the pressure still continuously rises in the same direction after removal is it considered a gas accumulation risk. Strong oil flow risk assessment uses a dual-evidence condition, requiring continuous baffle deflection and increased contact critical jitter to occur simultaneously and at the same time. Figure 4 The superimposed curves of pressure and oil temperature transitions within the two windows are presented. External electrical disturbance segments aligned with SC and TripSeq are marked with shaded areas for elimination. Short-term small drop points that occur during the nighttime window are also marked with symbols. Figure 4 The corresponding calculation results are Win1 event indices of 204 and 205. After removing the disturbance segment, the number of samples used for monotonicity calculation is 598. The net increase in Win1 is approximately 0.47 kPa, which is less than... The 0.8 kPa of _min and the existence of SC and TripSeq alignment events make the proof by contradiction unsatisfactory, therefore the risk of gas accumulation does not hold. Figure 4 The Win2 short-term pullback point index is given as 240, with a pullback amplitude of approximately 0.02 kPa and a duration of 1 second, satisfying the condition... For 3s and The suppression condition is 0.2 kPa, but the net increase in Win2 is about 0.30 kPa, which is still less than 0.8 kPa. Therefore, the risk of gas accumulation does not exist. The above results are consistent with the description of perturbation fingerprint elimination and short-term fallback suppression in this embodiment.

[0069] Step S3 involves session control and shadow writing. If a risk of gas accumulation or strong oil flow is detected in either window, self-calibration is disabled and the reason for the disabling is recorded. If neither risk is detected, the self-calibration session is entered and the action threshold tuning parameters are locked. This is read-only to prevent the threshold from being changed by external configuration during the session. Only read-only sampling is performed during the session, and the threshold compensation amount is calculated based on stable segment data. K and write it to the shadow parameter area. Simultaneously, the session time window Win and the sampling trust identifier ID are output for subsequent drift compensation constraints. This is to provide a verifiable demonstration of the dual-evidence condition for strong oil flow. Figure 5 The planar scatter plots of Theta and J are given, and the following are marked: and Threshold line. Figure 5 The corresponding calculation results show that the longest continuous duration for Win1 to simultaneously satisfy Theta not less than 8.0% and J not less than 0.65 is 0s, and the longest continuous duration for Win2 to simultaneously satisfy this dual evidence is also 0s. Therefore, the risk of strong oil flow is not established. Figure 5 The scattered points are concentrated in the lower left of the threshold line and there are no continuous synchronous over-limit segments. This is consistent with the field observations in this embodiment where Theta is stable at about 1.8% to 3.4% and J is stable at about 0.16 to 0.24, thus meeting the risk-free conditions for entering the self-calibration session.

[0070] Step S4 involves multi-time period consistency and write-back control. The system requires that sampling time periods under at least two different oil temperature transition states be provided. K, and K changes in the same direction and satisfies

[0071]

[0072] Only when satisfied K write update action threshold tuning parameter Otherwise, keep the original settings. After writing the update, store Win and ID, and limit the threshold compensation amount for the next write update to not exceed the maximum step size StepMax determined by ID.

[0073] To verify the above logic, data windows under two different oil temperature transition states of the same main transformer were selected for comparison. Each window was 600 seconds long. Representative segments within the windows are shown in the table for verifiable review. The complete sequence was sampled in 1-second increments and is traceable in the log. The first segment was the daytime heating phase, where the oil temperature jumped from 45℃ to 62℃. The second segment was the nighttime cooling phase, where the oil temperature dropped from 52℃ to 38℃. The timestamps for short-circuit current events and trip sequences were obtained from the protection device's SOE and waveform recording system. Disturbance segments were excluded from the pressure monotonicity assessment before drawing conclusions.

[0074]

[0075] As shown in Table 1, a pressure spike and subsequent drop occurred around 10:15:24.200 in Win1. This is a transient disturbance fingerprint aligned with the time of the short-circuit current event and trip sequence information. After removing this segment according to the disturbance fingerprint exclusion logic, the remaining pressure sequence still maintains an upward trend. In Win2, there is a short-term reverse fluctuation, with the drop lasting 1 second and an amplitude of 0.02 kPa, which is less than... No more than Even after applying short-term reverse fluctuation suppression measures, it is still determined that there will be no pullback. Combined with... Figure 4The elimination and suppression effects shown indicate that neither window triggered the risk of gas accumulation under the conditions of proof by contradiction and the constraint of net increase. Combined with... Figure 5 The planar distributions of Theta and J shown in the figure indicate that neither window triggered the dual evidence condition for strong oil flow. Therefore, neither type of risk is established, and the system allows entry into the self-calibration session and locks the window. .

[0076] The threshold compensation amount is calculated using an engineering-verifiable example. The original action threshold setting parameter is expressed as the equivalent threshold angle of the baffle action, initially... The value is 12.0%. During the session, the deviation between the equivalent trigger probability and the actual deflection distribution is statistically obtained from the stable segment, leading to the suggested threshold compensation amount. K is negative to suppress false alarms. Win1 calculates... K1 is -0.58%, written to the shadow parameter area. A sampled trusted identifier ID1 is generated, and the session time window is recorded from 10:12:00 to 10:22:00. Win2 calculates... K2 is -0.62%, written to the shadow parameter area and a sampled trusted identifier ID2 is generated. The session time window is recorded as 02:36:00 to 02:46:00. The two compensation quantities are in the same direction and satisfy the consistency constraint. The relative difference is calculated as follows:

[0077]

[0078] To meet the requirement of setting epsilon to 3% in this embodiment, the system further adds a 600s stable segment to the same night window for recalculation. K2 is corrected to -0.59%, at this point

[0079]

[0080] Figure 6 Displayed in a grouped columnar format K1, K2 original value and The K2 correction value is shown in the figure, along with the determination information for a consistency threshold epsilon of 3%. Figure 6 The corresponding calculation results show K2 original value and The relative difference of K1 is approximately 6.45%, which is greater than 3%, therefore consistency is not achieved; the original tuning should be maintained. The initial value was 12.0%. After recalculation and correction, the relative difference was approximately 1.69%, less than or equal to 3%, thus consistency was achieved, allowing for a write-back update and yielding a new threshold. _new is 11.42%. The session output is written to both Win and ID for subsequent drift compensation constraints, and the maximum step size StepMax is determined to be 0.8% based on ID, thus limiting the next update's (| K|) should not exceed 0.8% to avoid threshold mutations that could lead to protection coordination risks.

[0081] The following is an example of the system log for the write and output process: 10:22:01 Records entry into the self-calibration session and locking. For read-only mode, the record at 10:22:03 is written to the shadow parameter area. K1 and ID1 are marked with Win1, and the record is written to the shadow parameter area at 02:46:05. K2 and ID2 are marked with Win2. At 02:46:10, the consistency check is recorded and written back. _new is 11.42%, and 02:46:11 records the output session time window and sampling reliability identifier for subsequent drift compensation constraints. During routine daily inspections, maintenance personnel only need to confirm that A and T are both 0 and the protection has not activated; online correction of threshold drift can be completed without manual inspection.

[0082] The results are compared below: Before calibration, similar gas relays at this site experienced two false alarms within 12 hours of rapid oil temperature rise, with a false alarm rate of approximately 0.33 times per unit per month. After calibration, after 30 days of continuous operation, the system log showed zero false alarms. Furthermore, in one external short-circuit current event, the risk of gas accumulation was not triggered due to disturbance fingerprint removal, and the strong oil flow dual-evidence judgment remained invalid. This significantly reduced false alarms without compromising fault sensitivity. By using the output session time window and sampling credibility flag, subsequent drift compensation updates can be limited to a credible step range, avoiding protection coordination risks caused by threshold mutations.

[0083] Example 2:

[0084] The main transformer protection system of a 220kV substation was deployed. The same type of gas relay at this station had been in operation for three years. Operation and maintenance records showed a slow downward drift trend in the flow velocity damper's action threshold over the past six weeks. Disassembly and comparison revealed fatigue attenuation of the damper's return spring, leading to a slight increase in deflection angle under the same oil flow conditions and inducing increased critical contact jitter. The system uses online self-calibration to correct threshold drift during planned maintenance intervals, but a freeze was triggered during a suspected strong oil flow risk event. Subsequently, a safe write-back was completed through a sampling trusted identifier recovery process and adaptive annealing cooling logic, forming a complete timeline closed loop spanning several weeks.

[0085] The structure and version management of the shadow parameter area are as follows: The action threshold tuning parameter is denoted as... Both exist in the main parameter area and the shadow parameter area is used to carry the threshold compensation amount calculated during the session. K and its corresponding version number V. The shadow parameter is divided into two states: non-switchable draft and switchable version. The draft is used to retain candidate compensation quantities when trust is insufficient, but write-back is not allowed. The switchable version is used to form write-back candidates when trust reaches a preset level. The system generates a sampled trust identifier ID for each self-calibration session and implements hierarchical write control in the shadow parameter area. When the trust level reaches L3, the switchable version V is written and incremented by 1. When the trust level is L2, only the non-switchable draft is written. When the trust level is below L2, the formation of drafts or versions is prohibited and the session is exited.

[0086] Step S1 involves session initiation and risk assessment input acquisition. Following Example 1, the system completes risk assessment input and dual-evidence assessment input, forming a risk state sequence rt within each session window. rt = 0 indicates that neither gas accumulation risk nor strong oil flow risk is valid, while rt = 1 indicates that either risk is valid. Step S2 involves credibility score calculation and credibility level mapping. The system counts the number of risk state flips within each session window. And calculate the stable duration. With window length Simultaneously calculate the risk margin for gas accumulation. Risk margin with strong oil flow The credibility score C is calculated using the following formula.

[0087]

[0088] Where alpha is 0.35, The sampling period is 1 second and the duration is 600 seconds. The trust level mapping adopts the example threshold range: C not less than 0.60 is mapped to L3, C between 0.35 and 0.60 is mapped to L2, and C less than 0.35 is mapped to L1. Switchable versions are prohibited. Figure 7 The correspondence between the confidence scores C and the threshold bands for win1 to win3 is given. Figure 7 The corresponding calculation results are: win1 has a C value of 0.30 and is mapped to L1 with writes disabled; win2 has a C value of 0.14 and is mapped to L1 with writes disabled; win3 has a C value of 0.85 and is mapped to L3 with the ability to form a switchable version V7. Figure 7 The threshold band shown intuitively demonstrates the inhibitory effect of the number of risk flips and the minimum margin term on the credibility score, thereby supporting the shadow parameter differentiation level writing control logic.

[0089] Step S3 involves hierarchical writing and draft version formation. The system will calculate the threshold compensation amount during the session. K is written to the shadow parameter area and the ID and Win are recorded. When the trust level is L3, a switchable version is formed and marked as V. When the trust level is L2, a non-switchable draft is formed and write-back is not allowed. Step S4 is the freeze operation and write-back prohibition. If the risk of gas accumulation or strong oil flow is detected again during the self-calibration session, the freeze operation is performed. Freezing includes version sealing and write-back prohibition. Version sealing marks the most recently formed switchable version in the shadow parameter area as unwritable and records the corresponding session time window. Write-back prohibition prevents the switchable version from being written to the main parameter area during the subsequent preset cooling period. The freeze continues until a sampling credibility identifier satisfying L3 is regenerated. Version sealing further includes sealing reason binding and evidence snapshot recording. Sealing reason binding marks the risk type triggering the freeze as either gas accumulation risk or strong oil flow risk and stores it in association with the version number V. Evidence snapshot recording synchronously records the cavity pressure change characteristic P_feat, the baffle deflection characteristic Theta_feat, and the contact critical jitter characteristic at the time of freezing. . Figure 8 The freeze trigger point and three types of evidence snapshots are displayed on the same screen. The risk status segment is represented by a background color band and the freeze trigger point is marked at t=430s. Figure 8 The corresponding calculation results show that around t=430s, Theta reaches 9.2% and lasts for 6s, and J reaches 0.71 and lasts for 6s, satisfying the condition that short-term synchronization of dual evidence causes rt to jump from 0 to 1, triggering a freeze, and evidence snapshots are recorded simultaneously. The mutation amount within 1 second is 0.12 kPa and the average increase within 10 seconds is 0.18 kPa. Figure 8 This indicates that the freeze was not triggered by a single quantity error, but by an event that caused the deflection and jitter to exceed the limits simultaneously and was accompanied by stress characteristics, thus meeting the traceability requirements of binding the reason for version sealing and recording evidence snapshots.

[0090] To illustrate the timeline spanning several weeks, this embodiment selects three session windows (win1 to win3) and a session segment that triggered a freeze event as computable data support. Win1 occurred during the inspection window on Monday night of week 1, and the rt statistical summary shows that the value was 0 for the first 560 seconds and then flipped twice from 0 to 1 and back to 0 in the last 40 seconds. It is 2. It is 570s. It is 0.80. The value is 0.60. Win2 occurred during the morning rush hour on Wednesday of the second week. Due to the combined effects of load fluctuations and leaf spring fatigue, rt exhibited multiple fluctuations, with a value of 0 for 540 seconds and 1 for 30 seconds interspersed, and 3 flips occurring. It is 3. It is 540s. It is 0.55. The value was 0.40. This occurred late Friday night during the third week of Windows 3. After the system underwent cooling constraints and re-entered the self-calibration session, the rt value remained at 0 for 600 seconds without any flipping. =0, For 600 seconds, It is 0.95. The confidence score is 0.85. The confidence score C for each window and the shadow area writing results are shown in Table 2.

[0091] Table 2: Credibility and Shadow Differentiation Level Write Results from Win1 to Win3:

[0092] Session window RT Statistical Summary C value Trust level Shadow area results Win1 Monday of Week 1 After 560 seconds, a brief period of 1 appeared and the price fell back. 2 600s 570s 0.80 0.60 0.30 L1 Write prohibited Win2 Week 2 Wednesday 0 is the main component, mixed with dispersed 1 and shaking. 3 600s 540s 0.55 0.40 0.14 L1 Write prohibited Win3, Friday of Week 3 No flips throughout the entire process. 0 600s 600s 0.95 0.85 0.85 L3 Switchable version V7

[0093] The C value in Table 2 can be calculated item by item according to the formula. Taking Win3 as an example, =0, and Equal, min If it is 0.85, then

[0094]

[0095] A value of C not less than 0.60 is mapped to L3, thus allowing for switchable versions. In contrast, in Win1, min... It is 0.60. and The ratio is 0.95, and If it is 2, then

[0096]

[0097] Values ​​below 0.35 enter L1 and writes are prohibited, reflecting the suppressive effect of risk state reversal on credibility and the limiting effect of the minimum margin constraint. This mapping relationship is in Figure 7 The threshold bands are presented intuitively in the middle.

[0098] The freeze and evidence snapshot occurred during a suspected strong oil flow risk window on Thursday of Week 2. At the start of this window, the system had created a switchable version V6 based on a short-term stable period and was preparing to write it back after the session ended. However, at 430 seconds into the session, a short-term synchronous occurrence of the two strong oil flow evidence conditions was detected, causing rt to jump from 0 to 1 and triggering a freeze. The system performed version sealing, marking V6 as unwritable and binding the sealing reason as strong oil flow risk. Simultaneously, the session time window was recorded as 21:00:00 to 21:10:00 on Thursday of Week 2, and an evidence snapshot was recorded. The pressure change characteristics P_feat recorded in the evidence snapshot were a pressure surge of 0.12 kPa within 1 second and a 0.18 kPa increase in the average pressure within 10 seconds, and baffle deflection characteristics... The contact critical jitter characteristic lasted for 6 seconds and was 9.2%. The duration was 0.71 seconds, which met the synchronicity characteristics of suspected strong oil flow. After freezing, drafts that had not yet reached a switchable state in the same session were discarded, and a write-back was initiated to prevent entry into the cooling period. Figure 8 The synchronization between the aforementioned freeze trigger points and evidence snapshots is visualized, making it easier to directly align time and feature quantities during operation and maintenance review.

[0099] Table 3: Changes in Version Sealing, Evidence Snapshots, and Cooldown Periods

[0100] Event Number Version number Freeze or freeze? Reasons for sealing Write-back prohibition deadline Cooling period variation E1 Week 2, Thursday V6 yes Strong oil flow risk A 1-second mutation resulted in a 0.12 kPa increase, and the 10-second mean increased by 0.18 kPa. 9.2% lasting 6 seconds 0.71 lasts for 6 seconds L3 is generated for two consecutive sampling periods and rt is 0 for all of them. 24h extended to 72h E2, Monday of Week 3 none no none none none none Writeback ban lifted after reaching L3 72h shortened to 12h Friday of Week 3, E3 V7 no none none none none Satisfies L3 writeback capability Maintain for 12 hours

[0101] The cooldown period is determined using adaptive annealing logic. The initial cooldown time is 24 hours, with the annealing stepping rule being to add 12 hours for each failure, up to a maximum of 96 hours, and to subtract 6 hours for every two consecutive successes, with a minimum of 6 hours. After freezing on Thursday of Week 2, the trusted identifiers generated by the system in subsequent sampling cycles did not reach L3, causing the cooldown period to be gradually extended while maintaining write-back prohibition, eventually extending from 24 hours to 72 hours. Entering Monday of Week 3, the system generated L3 for two consecutive sampling cycles, and the risk assessment remained invalid, triggering successful annealing, shortening the cooldown period from 72 hours to 12 hours and removing the write-back restriction for subsequent versions. Figure 9 The process of dynamically extending and shortening the cooling period is displayed in the form of a stepped timeline. Figure 9 The corresponding calculation result is an initial cooling value of 24h. After freezing, the cooling is extended to 72h due to four consecutive failures to reach L3. Subsequently, the cooling is shortened to 12h and maintained until the win3 write-back stage after two consecutive attempts to reach L3 with rt all being 0. Figure 9 At the same time, the final write-back result K_th is updated from 12.0% to 11.40%, and the maximum step size StepMax is recorded as 0.8%, thereby avoiding the risk of protection coordination caused by one-time overcompensation.

[0102] The following is an example of the recovery and final write-back process: Master parameter area before freezing. The percentage is 12.0%, and the candidate compensation amount for V6 is... K is -0.55%, but write-back is prohibited due to freezing and sealing. After cooling and annealing, Win3 calculates... K is set to -0.60%. A switchable version V7 is created by writing this value to the shadow parameter area and updating K_th to 11.40% via the write-back window. Simultaneously, the sampling trust identifier ID3 is recorded, and the maximum step size for the next iteration is constrained to not exceed 0.8%, thus preventing excessive compensation for drift caused by spring fatigue. During the maintenance personnel's inspection at the end of the third week, they only need to verify the sealed records and write-back logs to complete the closed-loop confirmation; there is no need to disassemble the gas relay mechanism.

[0103] The results are compared below: Before the introduction of freeze and annealing, the site faced the risk of threshold erroneous writes when load fluctuations and temperature changes overlapped, potentially misinterpreting external disturbances as stable drift and leading to unreasonable reductions in protection thresholds. With the introduction of the tiered write and freeze write-back prohibition in this embodiment, even in the event of a suspected strong oil flow risk, the system can seal the version and record an evidence snapshot, avoiding erroneous write-backs. Simultaneously, adaptive annealing shortens the cooling time and improves availability after reliable recovery. Statistics show that before implementation, similar equipment experienced two false alarms within two weeks, requiring manual review. After implementation, no false alarms due to threshold writes occurred for four consecutive weeks, and the evidence snapshots of freeze events can be directly used for operational review, improving security and reducing the cost of handling false alarms.

Claims

1. A self-calibration and drift compensation method for a flow rate baffle in a transformer gas relay, characterized in that... include: Collect gas relay alarm status, trip status and cavity pressure changes, as well as transformer short circuit current events, trip sequence information and oil temperature transition information; When the cavity pressure continues to rise unidirectionally within a preset time, it is determined to be a risk of gas accumulation. When the flow rate baffle continues to deflect within a preset time and the contact point shows a critical increase in vibration, it is determined to be a risk of strong oil flow. If either risk is met, self-calibration is prohibited. When there is no risk of gas accumulation and no risk of strong oil flow, the self-calibration session is entered and the flow rate baffle action threshold setting parameter is locked. During the session, only read-only sampling is performed and the calculated threshold compensation amount is written to the shadow parameter area. When the threshold compensation amount remains consistent across multiple sampling periods, the threshold compensation amount is written into the update action threshold tuning parameter; otherwise, the original tuning is maintained. The session time window and sampling reliability identifier are output for subsequent drift compensation constraints.

2. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 1, characterized in that... The determination of the risk of gas accumulation satisfies the monotonicity of pressure change and the counter-evidence condition within a preset time. The monotonicity condition is that the cavity pressure rises in the same direction for multiple consecutive samplings without falling back. The counter-evidence condition is that no transient disturbance corresponding to a short-circuit current event or trip sequence is detected within the preset time. The determination of the risk of strong oil flow also satisfies the dual evidence condition within the preset time. The dual evidence condition is that the continuous deflection of the flow velocity baffle and the increase in the critical vibration of the contact point occur simultaneously and in a synchronous manner.

3. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 1, characterized in that... During the session, when the threshold compensation amount is written to the shadow parameter area, a sampling trust identifier is generated and hierarchical writing control is implemented. When the sampling trust identifier reaches the preset level, a switchable version is formed in the shadow parameter area. When the preset level is not reached, only a non-switchable draft is retained. Furthermore, when the risk of gas accumulation or strong oil flow is detected again during the self-calibration session, the switchable version is frozen and the draft is discarded.

4. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 1, characterized in that... The multiple sampling periods include at least two sampling periods under different oil temperature transition states, and the threshold compensation amount is required to change in the same direction and the difference does not exceed the consistency threshold. Only when this condition is met will the update action threshold tuning parameters be written. Write the updated storage session time window and the sampled trusted identifier, and limit the threshold compensation amount of the next write update to not exceed the maximum step size determined by the sampled trusted identifier.

5. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 2, characterized in that... The cavity pressure is continuously sampled within the preset time and short-term reverse fluctuations are suppressed. When an instantaneous drop occurs that does not exceed the preset duration and its amplitude is less than the preset drop threshold, it is still determined that there is no drop.

6. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 2, characterized in that... The counter-evidence conditions include disturbance fingerprint exclusion logic. When a short-circuit current event or trip sequence information is detected, the cavity pressure change segment that is aligned with its time within the preset time is marked as an external electrical disturbance segment and removed from the monotonicity judgment. The gas accumulation risk is only determined when the pressure continues to rise in the same direction after removal.

7. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 3, characterized in that... The sampling confidence identifier is generated by the risk discrimination stability within the self-calibration session. The risk discrimination stability includes that the risk of gas accumulation and the risk of strong oil flow remain false and do not undergo state reversal within a preset duration. The stability is then mapped to a confidence level. Only when the confidence level reaches a preset level is the switchable version allowed to be formed in the shadow parameter area.

8. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 3, characterized in that... The freeze operation when the risk of gas accumulation or strong oil flow is detected includes version sealing and write-back prohibition. Version sealing is to mark the most recently formed switchable version in the shadow parameter area as unwritable and record the corresponding session time window. Write-back prohibition is to prohibit the switchable version from being written to the update action threshold tuning parameter during the subsequent preset cooling period, until a sampling trust identifier that meets the preset level is regenerated.

9. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 8, characterized in that... The version sealing includes sealing reason binding and evidence snapshot recording. The sealing reason binding is to mark the risk type that triggers freezing as gas accumulation risk or strong oil flow risk and store it in association with the switchable version. The evidence snapshot recording is to record the cavity pressure change characteristics, flow baffle deflection characteristics and contact critical jitter characteristics at the moment of triggering freezing while recording the session time window.

10. The self-calibration and drift compensation method for the flow rate baffle of a transformer gas relay according to claim 8, characterized in that... The preset cooling period is determined using adaptive annealing logic. The adaptive annealing logic dynamically extends or shortens the cooling period based on the recovery process of the sampled trusted identifier. When the sampled trusted identifier does not reach the preset level within multiple consecutive sampling periods, the cooling period is extended. When the sampled trusted identifier reaches the preset level within multiple consecutive sampling periods and the risk judgment remains invalid, the cooling period is shortened.

Images