Oil-immersed transformer multi-sensor data intelligent compensation method

Abstract

The application discloses an oil-immersed transformer multi-sensing data intelligent compensation method, and relates to the technical field of electrical equipment monitoring; the method comprises the following steps: acquiring pressure sequence, oil temperature sequence and liquid level sequence under the same time axis, determining the slow change state of the oil temperature sequence and the liquid level sequence, constructing pressure reference information and determining pressure deviation information; identifying the wrong phase segment that is inconsistent with the slow change of the pressure step drop and the slow change of the oil temperature and liquid level according to the change difference between the pressure deviation information and the slow change state of the oil temperature and liquid level; performing backfill compensation on the pressure sequence corresponding to the wrong phase segment, generating a compensated pressure sequence, and determining a pressure release valve action representation quantity; determining a health degradation index and a fault prediction quantity based on the compensated pressure sequence, the slow change state of the oil temperature and liquid level, and the pressure release valve action representation quantity. The application can avoid the pressure step drop caused by the critical action of the pressure release valve from being misjudged as the real recovery of the internal pressure or the abnormality of the sensor, and improve the accuracy of the fault prediction and health management of the oil-immersed transformer.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment monitoring technology, and in particular to an intelligent compensation method for multi-sensor data of oil-immersed transformers. Background Technology

[0002] Oil-immersed transformers are widely used in new energy power plants, medium-voltage distribution networks, and industrial power transmission and distribution systems. Their operating status is closely related to internal pressure, oil temperature, and liquid level. Online monitoring devices typically use pressure sensors, oil temperature sensors, and liquid level sensors to continuously collect data from oil-immersed transformers and transmit the collected data to a backend system for data analysis, anomaly alarms, and health management. During the operation of an oil-immersed transformer, an increase in oil temperature causes oil expansion, and the liquid level changes slowly with the change in oil volume. The internal pressure is also affected by the buffering effect of the gas chamber, the thermal expansion of the oil, and the operation of the pressure relief valve. When the pressure relief valve is in the critical opening and reseating stage, a short-term pressure release may cause a step drop in the pressure sensor data, while the oil temperature and liquid level continue to change slowly due to thermal inertia, overall oil flow, and the buffering effect of the gas chamber. This results in a phase discrepancy where the pressure drops rapidly but the oil temperature and liquid level do not change synchronously. This situation makes it difficult for the backend system to determine whether the pressure drop is due to a true recovery of internal pressure, sensor malfunction, or a short-term pressure release caused by the critical operation of the pressure relief valve, thus affecting the accuracy of fault prediction and health management of oil-immersed transformers.

[0003] Existing online monitoring methods for oil-immersed transformers primarily focus on separately collecting, thresholding, displaying trends, or alarming for pressure, oil temperature, and liquid level. They typically treat pressure drops as a reduction in pressure risk or filter out sudden pressure changes as anomalies, lacking a dedicated mechanism for identifying and compensating for pressure step drops during the critical reseating process of pressure relief valves. Since pressure relief valve action is a short-term mechanical depressurization process, while oil temperature and liquid level are slowly changing parameters, the three types of data do not have completely consistent time responses. Simply relying on raw pressure values ​​or basic multi-sensor fusion can easily mask critical valve action events and may mistakenly delete fault prediction information that should be retained, leading to distortions in compensated pressure, health degradation indices, and maintenance recommendations. Therefore, a method is needed that can identify pressure step drops and slowly changing liquid temperature phase misalignment, and intelligently compensate for pressure data while retaining pressure relief valve action information. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies, such as the difficulty in distinguishing whether a pressure drop during the critical opening and reseating stages of a pressure relief valve is due to a true recovery of internal pressure, sensor malfunction, or short-term pressure relief of the valve components. This leads to distortion in the compensation of multi-sensor data (pressure, oil temperature, and liquid level) and affects the accuracy of fault prediction and health management. Therefore, this invention proposes an intelligent compensation method for multi-sensor data of oil-immersed transformers.

[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: A method for intelligent compensation of multi-sensor data in oil-immersed transformers includes: S1. Obtain the pressure sequence, oil temperature sequence, and liquid level sequence of the oil-immersed transformer at the same time axis; S2. Determine the slowly changing temperature and liquid state based on the oil temperature sequence and liquid level sequence, and construct a slowly changing pressure reference quantity based on the slowly changing temperature and liquid state. Obtain the pressure residual based on the deviation relationship between the pressure sequence and the slowly changing pressure reference quantity. S3. Based on the rapid decline of the pressure residual and the gradual change of the temperature liquid, determine the phase misalignment state between the rapid decline of pressure and the gradual change of temperature liquid, and extract the corresponding phase misalignment segment from the time axis. S4. Determine the pressure step fall characterization quantity based on the pressure residual change in the phase-misaligned segment, and perform equivalent backfilling on the pressure sequence based on the pressure step fall characterization quantity to generate a compensated pressure sequence. Determine the pressure relief valve action intensity quantity based on the pressure step fall characterization quantity and the phase misalignment degree of the phase-misaligned segment. S5. Calculate the health degradation index of the oil-immersed transformer based on the compensation pressure sequence, the gradual change state of the liquid temperature, and the action intensity of the pressure relief valve, and calculate the fault prediction quantity based on the changing trend of the health degradation index.

[0006] Preferably, the slowly changing temperature-liquid state is determined based on the oil temperature sequence and liquid level sequence, and a slowly changing pressure reference quantity is constructed based on the slowly changing temperature-liquid state. The pressure residual is obtained based on the deviation relationship between the pressure sequence and the slowly changing pressure reference quantity, including: A data window is constructed based on data adjacent to the current sampling time; The median of the oil temperature data and the liquid level data within the data window are calculated respectively, and the median is used as the oil temperature slow variable and the liquid level slow variable at the corresponding sampling time. The oil temperature slow variable and the liquid level slow variable together constitute the temperature and liquid slow change state. Obtain the pressure baseline coefficient, the first pressure influence coefficient, and the second pressure influence coefficient obtained by fitting historical stable operation segments of the transformer; Based on the pressure reference coefficient, the first pressure influence coefficient, the second pressure influence coefficient, the oil temperature slow change variable, and the liquid level slow change variable, a pressure slow change reference quantity is obtained; The pressure residual is obtained by subtracting the pressure sequence at the same sampling time from the slowly varying pressure reference.

[0007] Preferably, based on the rapid decline of the pressure residual and the gradual change of the temperature liquid, the phase misalignment state between the rapid pressure decline and the gradual temperature change is determined, and the corresponding phase misalignment segment is extracted from the time axis, including: Calculate the change in pressure residual, the gradual change in oil temperature, and the gradual change in liquid level between the current time and the previous time. Extract the positive part of the negative value of the pressure residual change, and extract the absolute value of the gradual change in oil temperature and the absolute value of the gradual change in liquid level. Adaptive normalization is performed on the positive part of the negative value of the pressure residual change, the absolute value of the oil temperature change, and the absolute value of the liquid level change by means of the median absolute deviation, thereby obtaining the pressure step drop intensity, the oil temperature change intensity, and the liquid level change intensity in sequence. Calculate the sum of the intensity of the gradual change in oil temperature and the intensity of the gradual change in liquid level, add the sum to the value of 1, and take the reciprocal to obtain the gradual change inhibition factor; Multiplying the pressure step fall intensity by the gradual suppression factor yields the phase misalignment intensity; The continuous time period during which the phase fault intensity exhibits a local peak and the pressure residual changes from decreasing to stabilizing or increasing is defined as a phase fault segment.

[0008] Preferably, the pressure step fall characterization quantity is determined based on the pressure residual change within the phase-shifted segment, and the pressure sequence is equivalently compensated based on the pressure step fall characterization quantity to generate a compensated pressure sequence, including: The difference between the pressure residual at the beginning of the phase-shifted segment and the minimum pressure residual within the phase-shifted segment is taken as the pressure step drop. The pressure recovery weight is obtained by dividing the difference between the pressure residual at the beginning of the phase misalignment segment and the pressure residual at the current moment by the pressure step drop amount. The pressure recovery weight is limited to a range that is not less than zero and not greater than one. For the sampling time belonging to the phase misalignment segment, the product of the pressure recovery weight and the pressure step drop amount at the corresponding time is added to the original pressure value to obtain the compensation pressure value within the segment. For sampling times that do not belong to the phase-shifted segment, the original pressure value is used as the external compensation pressure value, and the internal compensation pressure value and the external compensation pressure value together constitute the compensation pressure sequence.

[0009] Preferably, the pressure relief valve actuation intensity is determined based on the pressure step drop characteristic quantity and the phase misalignment degree of the phase misalignment segment, including: Calculate the arithmetic mean of the phase fault intensity at all sampling times within the phase fault segment; Multiply the pressure step drop by the arithmetic mean of the phase misalignment intensity to obtain the pressure relief valve action intensity corresponding to the phase misalignment segment.

[0010] Preferably, the health degradation index of the oil-immersed transformer is calculated based on the compensation pressure sequence, the gradual change state of the liquid temperature, and the intensity of the pressure relief valve operation, including: The pressure relief valve actuation intensity is mapped to the sampling time of the corresponding phase-out segment, and the value is set to zero at the non-phase-out segment time to obtain the pressure relief valve actuation intensity distributed on the time axis; Adaptive normalization based on median absolute deviation is performed on the compensation pressure sequence, the oil temperature variable, the liquid level variable, and the pressure relief valve action intensity, respectively, to obtain the compensation pressure normalization result, oil temperature normalization result, liquid level normalization result, and pressure relief valve action normalization result. Extract the portions greater than zero from the normalized results of the compensation pressure, the normalized results of the oil temperature, the normalized results of the liquid level, and the normalized results of the pressure relief valve action, respectively, as the corresponding positive degradation components for each item; The weights of each item are determined based on the relative dispersion of the compensation pressure sequence, the oil temperature variable, the liquid level variable, and the pressure relief valve action intensity in historical stable operating segments. The health degradation index is obtained by weighting and summing the positive degradation components of each item with their corresponding weights.

[0011] Preferably, the weights of each item are determined based on the relative dispersion of the compensated pressure sequence, the oil temperature buffer, the liquid level buffer, and the pressure relief valve action intensity in historical stable operating segments, including: Calculate the median absolute deviation of the compensation pressure sequence, the oil temperature slow variable, the liquid level slow variable, and the pressure relief valve action intensity in the historical stable operation segment; and use the ratio of the median absolute deviation of each item to the sum of the median absolute deviations of all items as the weight corresponding to that item.

[0012] Preferably, the fault prediction quantity is calculated based on the changing trend of the health degradation index, including: Calculate the difference between the health degradation index at the current time and the previous time, and divide the difference by the time interval between the current time and the previous time to obtain the rate of change of the degradation index; The rate of change of the degradation index is multiplied by the prediction time scale, and the product is added to the health degradation index at the current time to obtain the fault prediction amount. The prediction timescale is determined by the monitoring system data upload cycle, the equipment operation and maintenance cycle, or the prediction cycle selected by the user.

[0013] Preferably, after obtaining the fault prediction amount, the method further includes: Generate a critical action indication for the pressure relief valve based on the phase misalignment segment and the actuation intensity of the pressure relief valve; Transformer health status indicators are generated based on compensation pressure sequence, health degradation index, and fault prediction. Maintenance recommendations are generated when the number of predicted failures continues to increase relative to the health degradation index.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention acquires pressure, oil temperature, and liquid level sequences on the same time axis, constructs a slowly changing pressure reference using the slowly changing oil temperature and liquid level, and forms a pressure residual based on the deviation of the pressure sequence from the slowly changing pressure reference. This allows pressure changes caused by oil thermal expansion, gas chamber buffering, and slowly changing liquid level to be separated first. When the pressure relief valve is in the critical opening and reseating stage, it can extract the phase misalignment segment based on the rapid decrease of the pressure residual while the oil temperature and liquid level remain slowly changing. This avoids simply judging the pressure step drop caused by short-term pressure relief of the valve as a true recovery of internal pressure or sensor malfunction, thus improving the pertinence and accuracy of multi-sensor data compensation for pressure, oil temperature, and liquid level.

[0015] 2. This invention further performs equivalent backfilling on the pressure sequence based on the pressure step drop and pressure backfilling weight within the phase misalignment segment, generating a compensated pressure sequence. Simultaneously, it calculates the pressure relief valve actuation intensity based on the phase misalignment intensity and pressure step drop, enabling the pressure compensation process to correct data distortion caused by pressure step drop while preserving the critical actuation event of the pressure relief valve itself. The compensated pressure sequence, oil temperature gradient, liquid level gradient, and pressure relief valve actuation intensity are used together to calculate the health degradation index and fault prediction quantity, thereby enabling the backend to more accurately generate health status prompts, anomaly prompts, and maintenance suggestions, improving the reliability of fault prediction and health management for oil-immersed transformers. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating an intelligent compensation method for multi-sensor data of an oil-immersed transformer, provided in an embodiment of the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example: This example provides an intelligent compensation method for multi-sensor data of an oil-immersed transformer. See [link to example]. Figure 1Specifically, it includes the following steps: S1. Obtain the pressure sequence, oil temperature sequence, and liquid level sequence of the oil-immersed transformer at the same time axis; S2. Determine the slowly changing temperature and liquid state based on the oil temperature sequence and liquid level sequence, and construct a slowly changing pressure reference quantity based on the slowly changing temperature and liquid state. Obtain the pressure residual based on the deviation relationship between the pressure sequence and the slowly changing pressure reference quantity. S3. Based on the rapid decline of the pressure residual and the gradual change of the temperature liquid, determine the phase misalignment state between the rapid decline of pressure and the gradual change of temperature liquid, and extract the corresponding phase misalignment segment from the time axis. S4. Determine the pressure step fall characterization quantity based on the pressure residual change in the phase-misaligned segment, and perform equivalent backfilling on the pressure sequence based on the pressure step fall characterization quantity to generate a compensated pressure sequence. Determine the pressure relief valve action intensity quantity based on the pressure step fall characterization quantity and the phase misalignment degree of the phase-misaligned segment. S5. Calculate the health degradation index of the oil-immersed transformer based on the compensation pressure sequence, the gradual change state of the liquid temperature, and the action intensity of the pressure relief valve, and calculate the fault prediction quantity based on the changing trend of the health degradation index.

[0019] In one embodiment of the present invention, obtaining the pressure sequence, oil temperature sequence, and liquid level sequence of an oil-immersed transformer at the same time axis includes: This invention discloses an intelligent compensation method for multi-sensor data of an oil-immersed transformer, deployed in the data analysis unit of an online monitoring device for an oil-immersed transformer. The online monitoring device includes a pressure sensor for collecting internal pressure, an oil temperature sensor for collecting transformer oil temperature, a liquid level sensor for collecting transformer liquid level, a signal acquisition and control unit for receiving sensor signals, and a data processing unit for performing data compensation, anomaly identification, health degradation assessment, and fault prediction. The pressure sensor, oil temperature sensor, and liquid level sensor form a data stream with the same sampling period, either with the same sampling period or after resampling. The signal acquisition and control unit sends the collected data to the background data processing unit via RS-485 communication or other industrial communication methods. The background data processing unit performs buffering, verification, time alignment, and continuous analysis of the pressure data, oil temperature data, and liquid level data according to the sampling time, ensuring that each sampling time has a corresponding pressure value, oil temperature value, and liquid level value. Specifically, the sampling times formed during the continuous operation of the oil-immersed transformer are as follows: ; In the formula, This represents the i-th sampling time, where i represents the sampling time number and n represents the total number of sampling times currently involved in the calculation. The sampling time is determined by the data upload timestamp of the online monitoring device or the unified timestamp from the backend. If there is a slight offset in the upload times of different sensors, data under the same time axis is generated using nearest neighbor alignment or linear interpolation alignment based on a unified time axis. Under the same time axis, the pressure sequence, oil temperature sequence, and liquid level sequence are represented as follows: ; In the formula, This represents the internal pressure value at the i-th sampling time. The data comes from a pressure sensor, and the unit can be kilopascals or the same pressure unit as the pressure sensor output. This represents the oil temperature value at the i-th sampling time. The data comes from the oil temperature sensor and the unit can be degrees Celsius. The value represents the liquid level at the i-th sampling moment. The data comes from the liquid level sensor, and the unit can be millimeters, percentages, or the same liquid level unit as the output of the liquid level sensor. The pressure sequence is used to calculate the pressure residual, identify pressure step drop, and generate a compensation pressure sequence. The oil temperature sequence and liquid level sequence are used to construct the temperature and liquid slowly changing state and form a pressure slowly changing reference quantity, thereby ensuring that the three types of sensor data do not participate in the judgment in isolation, but form a calculable multi-sensor correspondence at the same sampling moment.

[0020] In one embodiment of the present invention, a slowly changing temperature-liquid state is determined based on an oil temperature sequence and a liquid level sequence, and a slowly changing pressure reference quantity is constructed based on the slowly changing temperature-liquid state. The pressure residual is obtained based on the deviation relationship between the pressure sequence and the slowly changing pressure reference quantity, including: After time alignment, the data processing unit performs gradual change extraction on the oil temperature and liquid level sequences to obtain a temperature-liquid gradual change state that reflects the thermal inertia, overall expansion, and buffering effect of the oil. It should be noted that the temperature-liquid gradual change state refers to the combined state formed by the gradual change extraction of the oil temperature and liquid level sequences. This means that it represents the slow change trend jointly formed by the thermal inertia of the transformer oil, changes in oil volume, and the buffering effect of the gas chamber. The temperature-liquid gradual change state is not a single oil temperature value or a single liquid level value, but rather a combination of slow oil temperature and slow liquid level changes. This is used to compare with rapid pressure changes, thereby determining whether pressure changes deviate from the slow change pattern that can be explained by oil temperature and liquid level.

[0021] Specifically, a data window is constructed centered on the current sampling time or based on neighboring sampling times preceding the current sampling time. It should be noted that a data window refers to a set of continuous and valid sampled data selected around the current sampling time, whose function is to provide a calculation basis for the slow changes in oil temperature and liquid level at the current sampling time. The data window is not limited to a fixed length, but is determined based on the online or offline analysis method, the sampling period, and the continuity of valid data; its purpose is to extract slow changing trends using neighboring sampled data and reduce the impact of isolated spikes on subsequent slow pressure reference values. The data window is represented as: ; In the formula, This represents the data window corresponding to the i-th sampling time. This represents the j-th sampling time within the data window, where j represents the sampling time number within the window. The data window represents the set of sampling sequence numbers adjacent to the i-th sampling time. The length of the data window can be determined based on the sampling cycle of the online monitoring device, the inertia of transformer oil temperature changes, and the background calculation refresh cycle, and can be adaptively adjusted using historical operating data. Furthermore, the data window consists of consecutive valid sampling points adjacent to the current sampling time. When the data processing unit performs offline analysis, the data window includes consecutive valid sampling points before and after the current sampling time. When the data processing unit performs online real-time analysis, the data window includes the current sampling time and consecutive valid sampling points before the current sampling time. If there are sampling points marked as communication abnormalities, over-range abnormalities, or missing sampling points within the data window, these sampling points are not included in the median calculation. If consecutive valid sampling points still exist after removing abnormal sampling points, the remaining consecutive valid sampling points are used to calculate the oil temperature buffer variable and liquid level buffer variable. If the consecutive valid sampling points are insufficient to form the median calculation conditions, the current sampling time does not perform phase error segment identification, or the oil temperature buffer variable and liquid level buffer variable from the previous valid sampling time are used as temporary buffer variables, and a temporary calculation mark is added to the current sampling time. Take the median of the oil temperature and liquid level data in the data window to obtain the oil temperature slow variable and liquid level slow variable. ; In the formula, This represents the slow variable of oil temperature at the i-th sampling time. This represents median operations. This represents the oil temperature value at the j-th sampling time within the data window. This indicates that the oil temperature value being calculated belongs to the data window at the i-th sampling time. The median is used because oil temperature has a strong slow change during normal operation, and the median can reduce the impact of single-point sampling glitches, instantaneous communication anomalies, and short-term sensor disturbances on the slow change trend of oil temperature, making the obtained slow oil temperature variable more suitable as the input of the slow pressure change reference. ; In the formula, This represents the slow change in liquid level at the i-th sampling time. This represents median operations. This represents the liquid level value at the j-th sampling time within the data window. The calculated liquid level value belongs to the data window at the i-th sampling time. The median is used because the liquid level is affected by the overall thermal expansion of the oil and the buffering effect of the gas chamber, and usually does not undergo synchronous abrupt changes with the short-term action of the pressure relief valve. Extracting the slow-changing liquid level variable using the median preserves the slow-changing trend of the liquid level and suppresses instantaneous jumps. The slow-changing oil temperature variable and the slow-changing liquid level variable together constitute the slow-changing temperature-liquid state. This slow-changing temperature-liquid state is subsequently used to establish a slow-changing pressure reference value and to determine whether a rapid pressure drop is asynchronous with changes in oil temperature and liquid level.

[0022] After obtaining the slowly changing temperature and liquid level state, the data processing unit constructs a slowly changing pressure reference quantity based on the slowly changing oil temperature and liquid level variables. It should be noted that the slowly changing pressure reference quantity refers to the internal pressure reference level that can be explained by the slowly changing oil temperature and liquid level variables, assuming no critical action of the pressure relief valve occurs. This slowly changing pressure reference quantity is not the pressure value directly collected by the pressure sensor, but rather a pressure benchmark fitted based on the slowly changing temperature and liquid level state, used to represent the normal influence of slowly changing factors such as oil thermal expansion, heat dissipation, and gas chamber compression on pressure. Specifically, historical stable operating segments are selected from the historical operating data of oil-immersed transformers. It should be noted that a historical stable operating segment refers to a continuous and valid data segment that can represent the normal operating relationship of the oil-immersed transformer, and it is used to fit the correlation coefficient in the slowly changing pressure reference value. This segment should exclude data intervals that have been identified as phase errors, have continuous missing data, have experienced pressure drop alarms, have pressure relief valve critical operation indications, or have sensor over-range anomalies, so that the fitted pressure reference relationship reflects the pressure change law under normal temperature and liquid slowly changing conditions as much as possible. The historical stable operation segment is a continuous and valid data segment selected from historical operation data. This continuous and valid data segment meets the following conditions: it is not identified as a phase-out segment; no pressure relief valve action prompts are recorded; no pressure drop alarms occur; the pressure residual does not show continuous step-down drops; the pressure sequence does not show over-range anomalies; and both the oil temperature and liquid level slow-change variables maintain continuous change. Specifically, the data processing unit first removes sampling intervals from historical data that have continuous missing values, communication anomalies, sensor over-range anomalies, or have been marked as critical pressure relief valve actions. Then, it selects operation segments from the remaining data where the pressure residual change direction is continuous and the oil temperature and liquid level slow-change variables do not undergo abrupt changes as historical stable operation segments. When the equipment is initially put into operation and lacks sufficient historical data, the data processing unit uses the first operation segment after commissioning that does not trigger an abnormal alarm and where the pressure sequence, oil temperature sequence, and liquid level sequence are all continuously valid as the initial stable operation segment. During subsequent operation, the historical stable operation segment is continuously updated using newly added normal operation data.

[0023] The data processing unit fits the pressure reference coefficient, the first pressure influence coefficient, and the second pressure influence coefficient using historical stable operating segments, and calculates the slowly varying pressure reference value according to the following formula: ; In the formula The pressure slowly varying reference quantity at the i-th sampling time is used to characterize the pressure level that can be explained by the slowly varying oil temperature and the slowly varying liquid level when the pressure relief valve has not been affected by the critical action of the pressure relief valve. This represents the pressure reference coefficient, which is obtained by fitting the pressure reference level from historical stable operating segments. This represents the first pressure influence coefficient, used to characterize the degree of influence of slowly changing oil temperature on slowly changing pressure reference quantity; This represents the second pressure influence coefficient, used to characterize the degree of influence of slowly changing liquid level on slowly changing pressure reference quantity; This represents the slow change in oil temperature at the i-th sampling time. This represents the slow-change pressure variation at the i-th sampling time. The above coefficients can be obtained through least-squares fitting, robust regression fitting, or condition-specific fitting of historical stable operating segments, enabling oil-immersed transformers with different capacities, oil tank structures, installation environments, and sensor calibration states to obtain slow-change pressure reference quantities adapted to their own operating characteristics. The reason for adopting this calculation method is that oil temperature and liquid level changes reflect slow-change factors such as transformer oil thermal expansion, heat dissipation process, and gas chamber compression, while the critical action of the pressure relief valve causes a short-term pressure drop. By first establishing a slow-change pressure reference quantity, the pressure portion that can be explained by the slow-change factors of temperature and liquid level can be separated from the portion of rapid pressure drop.

[0024] In one alternative implementation, the pressure reference coefficient, the first pressure influence coefficient, and the second pressure influence coefficient can be obtained by fitting the following formula: ; In the formula, This represents the set of sampling times corresponding to historical stable operation segments. This represents the r-th sampling time in a historical stable operation segment. This represents the internal pressure value at the r-th sampling time. This represents the slow change in oil temperature at the r-th sampling time. This represents the slow change in liquid level at the r-th sampling time. , and Represents the candidate coefficients to be fitted. This represents the coefficient combination that minimizes the sum of squared errors. This fitting method allows the slowly varying pressure reference value to match the historical stable operating relationship of the equipment. In the presence of a few historical outliers, the least squares fitting described above can be replaced with robust regression fitting to reduce the impact of outliers on the coefficients.

[0025] The deviation between the pressure sequence and the slowly varying pressure reference is calculated to obtain the pressure residual. It should be noted that the pressure residual refers to the deviation of the original pressure value relative to the slowly varying pressure reference. Physically, it represents the pressure change remaining after subtracting the slowly varying pressure components that can be explained by the slow variations in oil temperature and liquid level from the pressure sensing data. The pressure residual is used to highlight non-slowly varying pressure changes caused by short-term pressure relief valve depressurization, reseating hysteresis, or abnormal pressure sensing. It serves as the foundational data for subsequent identification of pressure step drops and phase misalignments. Pressure residuals are calculated using the following formula: ; In the formula, This represents the pressure residual at the i-th sampling time. This represents the original internal pressure value at the i-th sampling time. This represents the slowly varying pressure reference value at the i-th sampling time. The pressure residual is used as the subsequent identification target because it eliminates pressure components that can be explained by slow changes in oil temperature and liquid level, highlighting non-slowly varying pressure changes caused by short-term pressure relief valve depressurization, reseating lag, or abnormal pressure sensing. At this point, the slow oil temperature variable, the slow liquid level variable, and the pressure residual are all used as inputs for the subsequent phase misalignment identification step. The pressure residual is used to determine whether the pressure has dropped rapidly, while the slow oil temperature variable and the slow liquid level variable are used to determine whether the oil temperature and liquid level are still changing slowly.

[0026] In one embodiment of the present invention, based on the rapid decline of the pressure residual and the gradual change of the temperature liquid, the phase misalignment state between the rapid decline of pressure and the gradual change of temperature liquid is determined, and the corresponding phase misalignment segment is extracted from the time axis, including: To identify the phase discrepancy between rapid pressure drop and gradual temperature change, it's important to note that a pressure step drop refers to a short, rapid decrease in the pressure residual within adjacent or consecutive sampling times, distinct from gradual pressure changes caused by oil temperature or liquid level variations. In this embodiment, this pressure step drop is primarily used to characterize the discontinuous pressure drop caused by short-term pressure relief when the pressure relief valve is in the critical opening and reseating phase. The changes in pressure residual, oil temperature, and liquid level between the current sampling time and the previous sampling time are calculated respectively: ; In the formula, This represents the change in pressure residual at the i-th sampling time relative to the previous sampling time. This represents the pressure residual at the i-th sampling time. This represents the pressure residual at the (i-1)th sampling time. This change is used to characterize the direction and magnitude of the pressure residual's rise and fall between adjacent sampling times. When the value is negative and the absolute value is large, it indicates that the pressure residual is showing a rapid downward trend; ; In the formula, This represents the gradual change in oil temperature at the i-th sampling time relative to the previous sampling time. This represents the slow variable of oil temperature at the i-th sampling time. This represents the slow change in oil temperature at the (i-1)th sampling time. This change is used to characterize whether the slow change in oil temperature has changed significantly. If the change in oil temperature is small, it indicates that the oil temperature is still in a slow change state. ; In the formula, This represents the gradual change in liquid level at the i-th sampling time relative to the previous sampling time. This represents the slow change in liquid level at the i-th sampling time. This represents the slow change in liquid level at the (i-1)th sampling time. This change is used to characterize whether the slow change in liquid level has changed significantly. If the liquid level changes slowly and the change is small, it indicates that the liquid level is still affected by the overall expansion of the oil and the buffering effect of the gas chamber, and thus remains slow.

[0027] Since this invention focuses on the rapid decline of the pressure residual rather than its increase, the data processing unit extracts the positive portion of the inverse of the pressure residual change to obtain the pressure decline component. It should be noted that the pressure decline component refers to the change in pressure residual after retaining only the direction of its decline. When the pressure residual increases or remains constant, the pressure decline component is zero; when the pressure residual decreases, the pressure decline component reflects the magnitude of the decrease. The pressure drop component is calculated using the following formula: ; In the formula, This represents the pressure drop component at the i-th sampling time. This represents the change in pressure residual. This indicates that the maximum value is taken during the calculation. A value of 0 indicates that the pressure drop component does not participate in the phase reversal enhancement when the pressure residual does not decrease. The reason for using this calculation method is that the critical opening and reseating stages of the pressure relief valve usually show a short-term decrease in the pressure residual. If the pressure residual increases or remains unchanged, it should not be taken as positive evidence of a step drop in pressure. To avoid insufficient compatibility between different devices due to fixed thresholds, the data processing unit uses the median absolute deviation to adaptively normalize the absolute values ​​of the pressure drop component, the gradual change in oil temperature, and the gradual change in liquid level. In this embodiment, the median absolute deviation is defined as: ; In the formula, This represents the median absolute deviation of sequence X, where X represents the data sequence used in the robust dispersion calculation. This represents the m-th data value in sequence X, where m represents the data index within the sequence. This represents the median of sequence X. This represents the absolute deviation of the m-th data value from the median. The reason for using the absolute deviation of the median is that it is not sensitive to a small number of abrupt changes and outliers, making it suitable for adaptive scaling normalization of online monitoring data of oil-immersed transformers.

[0028] The pressure step fall intensity is calculated according to the following formula: ; In the formula, This represents the pressure step drop intensity at the i-th sampling time. This represents the pressure drop component at the i-th sampling time. This represents a sequence composed of pressure drop components at each sampling time within the current analysis period. This represents the median absolute deviation of the pressure retracement component sequence. This represents a very small positive number used to avoid division by zero. Its value can be determined by the calculation precision of the data processing unit or the sensor resolution, and it maintains the same meaning in all denominator protections in this embodiment. This very small positive number serves only as a calculation protection item to prevent division by zero and suppress abnormal numerical amplification. The very small positive number can be determined based on the floating-point calculation precision of the data processing unit, the minimum resolution of the pressure sensor, the minimum resolution of the oil temperature sensor, or the minimum resolution of the liquid level sensor, or it can be uniformly assigned a value by the background system during data initialization; the very small positive number remains consistent within the same analysis cycle. The reason for using this calculation method is that the pressure drop component needs to be measured relative to the pressure fluctuation scale of this equipment in this time period. After normalization, the relative strength of the rapid pressure drop under different operating conditions can be compared. The intensity of a gradual change in oil temperature is calculated using the following formula: ; In the formula, This represents the intensity of the gradual change in oil temperature at the i-th sampling time. This represents the absolute value of the gradual change in oil temperature at the i-th sampling time. This represents a sequence consisting of the absolute values ​​of the gradual changes in oil temperature at each sampling time within the current analysis period. This represents the absolute deviation of the median of the sequence. This represents an extremely small positive number used to avoid a denominator of zero. The reason for using this calculation method is that only when the oil temperature change is not synchronously significant does the rapid drop in pressure residual better reflect the physical behavior of the pressure relief valve's short-term pressure relief and the slow change in oil temperature. The intensity of a gradual change in liquid level is calculated using the following formula: ; In the formula, This represents the intensity of the gradual change in liquid level at the i-th sampling time. This represents the absolute value of the gradual change in liquid level at the i-th sampling time. This represents a sequence consisting of the absolute values ​​of the gradual changes in liquid level at each sampling time within the current analysis period. This represents the absolute deviation of the median of the sequence. This represents an extremely small positive number used to avoid a denominator of zero. The reason for using this calculation method is that the liquid level is affected by the overall thermal expansion of the oil and the buffering effect of the gas chamber. If the liquid level does not change abruptly but the pressure residual drops rapidly, it further indicates that the pressure drop should not be directly interpreted as a true recovery of the overall pressure of the oil system.

[0029] After obtaining the intensity of the pressure step drop, the intensity of the gradual change in oil temperature, and the intensity of the gradual change in liquid level, the gradual change inhibition factor is calculated: ; In the formula, Represents the slowly varying inhibition factor at the i-th sampling time. Indicates the intensity of a gradual change in oil temperature. This indicates the intensity of the gradual change in liquid level. The value of 1 is used to ensure that the denominator is not less than 1 and that the gradual change inhibition factor is calculable. The reason for using this calculation method is that when oil temperature and liquid level also change significantly and synchronously, the pressure drop may come from the overall change in operating conditions rather than the short-term pressure relief valve, in which case the gradual change inhibition factor decreases; when oil temperature and liquid level remain gradually changing, the gradual change inhibition factor is larger, making it easier for a rapid pressure drop to form evidence of phase discrepancy.

[0030] The phase fault intensity is calculated according to the following formula: ; In the formula, This represents the phase misalignment intensity at the i-th sampling time. This indicates the intensity of the pressure step drop. This represents the gradual change suppression factor. It's important to note that the phase misalignment intensity refers to the degree of inconsistency between the rapid decline in pressure residual and the continued gradual changes in oil temperature and liquid level. A higher phase misalignment intensity indicates a more pronounced short-term pressure relief characteristic in the pressure change, and a greater inability to interpret the pressure drop as a synchronous change in the overall operating state due to oil temperature and liquid level variations. Therefore, phase misalignment intensity is used to distinguish between ordinary pressure fluctuations and phase misalignment caused by the critical reseating hysteresis of the pressure relief valve. The reason for using this calculation method is that the key manifestation of the critical reseating hysteresis of the pressure relief valve is not a single pressure drop, but rather a rapid decline in pressure residual while the gradual changes in oil temperature and liquid level do not change in the opposite direction synchronously. Therefore, multiplying the pressure step drop intensity by the gradual change suppression factor allows the phase misalignment intensity to centrally reflect the degree of inconsistency between the rapid pressure drop and the gradual changes in temperature and liquid level.

[0031] The data processing unit extracts phase-shifted segments from the time axis based on the phase-shifted intensity sequence and the pressure residual variation trend. Specifically, when the phase-shifted intensity forms a local peak within a local time range, and the pressure residual transitions from decreasing to stabilizing or increasing within that local time range, this continuous time period is defined as a phase-shifted segment. It should be noted that a phase-shifted segment refers to a data interval on the time axis where there is a continuous inconsistency between a step-down pressure drop and a gradual change in temperature. This segment includes the process of the pressure residual rapidly decreasing to stabilizing or increasing, and is used to determine the effective range of subsequent equivalent pressure compensation. A phase-shifted segment is not an arbitrary abnormal time period, but rather a data interval jointly defined by the local peak of the phase-shifted intensity and the pressure residual variation trend. The phase-shifted segment is represented as: ; In the formula, This represents the k-th phase misalignment segment, where k represents the segment number. This indicates the start time of the k-th phase misalignment segment. This indicates the end time of the k-th phase misalignment segment. The local peak value can be determined by comparing adjacent sampling points, identifying the maximum value of the sliding window, or judging the peak duration. The transition of the pressure residual from decreasing to stabilizing or rising can be determined by the change in pressure residual value at adjacent sampling times gradually approaching zero or turning positive. The reason for using this segment extraction method is that the critical action of the pressure relief valve usually includes a short-term opening, pressure relief and falling back, and reseating and closing process. The pressure residual curve shows a rapid decrease followed by stabilization or rising. Extracting this process as a continuous segment can provide a clear compensation range for subsequent pressure equivalent replenishment.

[0032] In one embodiment of the present invention, a pressure step fall characterization quantity is determined based on the pressure residual change within the phase-misaligned segment, and the pressure sequence is equivalently compensated based on the pressure step fall characterization quantity to generate a compensated pressure sequence. The pressure relief valve actuation intensity quantity is determined based on the pressure step fall characterization quantity and the phase misalignment degree of the phase-misaligned segment, including: For each identified phase misalignment segment, the data processing unit calculates a pressure step fall characteristic quantity. In this embodiment, the pressure step fall characteristic quantity is represented by the pressure step fall amount. It should be noted that the pressure step fall amount refers to the difference between the pressure residual at the beginning of the phase misalignment segment and the lowest pressure residual within that segment, used to characterize the short-term fall amplitude caused by the critical action of the pressure relief valve on the pressure sensing data. This quantity does not represent the degree of permanent pressure reduction inside the transformer, but is used to calculate the equivalent compensation amount and the pressure relief valve action intensity. The pressure step drop is calculated using the following formula: ; In the formula, This represents the pressure step drop in the k-th phase misalignment segment. This represents the pressure residual at the start time of the k-th phase misalignment segment. This represents the minimum pressure residual within the k-th phase-shifting segment. The minimum pressure residual is obtained according to the following formula: ; In the formula, This represents the minimum pressure residual within the k-th phase-shifting segment. This represents the pressure residual at the i-th sampling time. This indicates that the i-th sampling time belongs to the k-th phase-out segment. The reason for using the pressure step drop as the pressure step drop characteristic is that the pressure residual at the beginning of the phase-out segment represents the residual level before the valve's critical action, and the minimum pressure residual within the phase-out segment represents the minimum residual level formed after short-term pressure relief. The difference between the two can directly characterize the drop in pressure sensing data caused by the critical action of the pressure relief valve. Next, the pressure recovery weights for each sampling moment within the phase-misaligned segment are calculated. It should be noted that the pressure recovery weight refers to the proportion of pressure relief at each sampling moment within the phase-misaligned segment relative to the pressure step drop, and it is used to control the pressure recovery amplitude at different sampling moments. A larger pressure recovery weight indicates that the sampling moment is more significantly affected by the pressure step drop, and the amount of pressure to be recovered is greater; a smaller pressure recovery weight indicates that the sampling moment is less affected by the pressure step drop. The pressure recovery weight is calculated according to the following formula: ; In the formula, This represents the pressure recovery weight at the i-th sampling time within the k-th phase misalignment segment. This represents the pressure residual at the start of the k-th phase misalignment segment. This represents the pressure residual at the i-th sampling time within the phase-out segment. This represents the pressure step drop in the k-th phase misalignment segment. This represents a very small positive number used to avoid a denominator of zero. This calculation method can characterize the degree of pressure residual decrease that has occurred at the current sampling time relative to the start time of the phase-shifted segment. By calculating the ratio of this to the pressure step drop, the pressure relief ratio corresponding to that sampling time can be obtained. This allows the compensation amount to change smoothly with the pressure residual decrease process within the segment, rather than performing a coarse compensation of the same magnitude for the entire segment.

[0033] To prevent individual outliers from causing the pressure recovery weight to exceed a reasonable range, the pressure recovery weight is subject to a limit: ; In the formula, This indicates the weighting of pressure recovery after the price limit is applied. This indicates the pressure recovery weight before the limit is set. The reason for using this limit method is that the pressure recovery weight essentially represents the proportion of the pressure step drop at the current sampling time. The reasonable range should be no less than zero and no greater than one. After the limit is set, the compensation pressure can be avoided from being over-compensated or reversed due to abnormal residual points. For sampling moments within out-of-phase segments, the data processing unit generates intra-segment compensation pressure values ​​based on the pressure recovery weight after amplitude limiting and the pressure step drop amount: ; In the formula, This represents the compensation pressure value at the i-th sampling time. This represents the original internal pressure value at the i-th sampling time. This represents the pressure recovery weight after amplitude limiting at the ith sampling time within the k-th phase-shifted segment. This represents the pressure step drop in the k-th phase misalignment segment. This indicates that the i-th sampling time belongs to the k-th phase-out segment. The short-term pressure drop within the phase-out segment is mainly caused by the critical action of the pressure relief valve. Directly using this as the true internal pressure recovery would lead to compensation distortion. Therefore, the step drop is equivalently superimposed back to the original pressure value according to the pressure recovery weight, which can restore the pressure continuity in the pressure sequence that is obscured by the short-term pressure relief disturbance of the valve. For sampling moments that do not belong to out-of-phase segments, the data processing unit keeps the original pressure value unchanged: ; In the formula, This represents the compensation pressure value at the i-th sampling time. This represents the original internal pressure value at the i-th sampling time. This indicates that the i-th sampling time does not belong to the k-th phase-disrupted segment. Non-phase-disrupted segments are not identified as pressure step drop intervals caused by the critical reseating hysteresis of the pressure relief valve, and their original pressure data should not be unfoundedly supplemented. Therefore, the compensated pressure values ​​within and outside these segments together constitute the compensated pressure sequence. It should be noted that the compensated pressure sequence refers to the pressure data sequence formed after equivalently supplementing the original pressure values ​​within the phase-disrupted segments and keeping the original pressure values ​​within the non-phase-disrupted segments unchanged. This sequence is used to replace the original pressure sequence in subsequent health degradation index calculations and status displays. Its purpose is to avoid misinterpreting the pressure step drop caused by short-term pressure relief valve decompression as a true recovery of internal pressure.

[0034] While generating the compensation pressure sequence, the critical action event of the pressure relief valve itself is retained to avoid erasing fault prediction information during the compensation process. Specifically, the data processing unit calculates the arithmetic mean of the phase error intensity at all sampling times within the phase error segment: ; In the formula, This represents the average phase fault intensity of the k-th phase fault segment. This represents the number of sampling points within the k-th phase-disrupted segment. The intensity of the phase misalignment at the i-th sampling time indicates that the reliability of the critical action of the pressure relief valve depends not only on a single peak value, but also on the overall duration of the inconsistency between the rapid pressure drop and the gradual temperature change within the phase misalignment segment. The average value can stably characterize the overall phase misalignment of the segment. The actuation strength of the pressure relief valve is calculated according to the following formula: ; In the formula, This represents the intensity of the pressure relief valve action corresponding to the k-th phase misalignment segment. This represents the pressure step drop in the k-th phase misalignment segment. This represents the average phase misalignment intensity of the k-th phase misalignment segment. It should be noted that the pressure relief valve actuation intensity quantity refers to a segment-level state quantity used to characterize the strength of the critical actuation event of the pressure relief valve. It simultaneously includes the pressure step drop amplitude and the persistence reliability of the phase misalignment phenomenon. This quantity is not the pressure value directly measured by the pressure sensor, but rather an event characterization quantity calculated from the phase misalignment segment. It is used to retain the contribution of the pressure relief valve's critical actuation to fault prediction and health management after pressure compensation. The reason for using this calculation method is that the pressure step drop amplitude reflects the pressure decrease caused by the critical actuation of the pressure relief valve, and the average phase misalignment intensity reflects the degree of inconsistency between this decrease amplitude and the gradual change in oil temperature and level. Multiplying the two can simultaneously retain the pressure relief amplitude and the reliability of the phase misalignment, allowing the pressure relief valve actuation intensity quantity to be used both for event indication and for subsequent health degradation index calculation.

[0035] In one embodiment of the present invention, a health degradation index of the oil-immersed transformer is calculated based on the compensation pressure sequence, the gradual change state of the liquid temperature, and the intensity of the pressure relief valve operation. A fault prediction quantity is then calculated based on the changing trend of the health degradation index, including: During the health degradation index calculation phase, the data processing unit maps the pressure relief valve actuation intensity to the sampling time of the corresponding out-of-phase segment, and assigns a value of zero at non-out-of-phase segment times, thus obtaining the pressure relief valve actuation intensity distributed along the time axis. It should be noted that the pressure relief valve actuation intensity is a time-series state quantity obtained by mapping the segment-level pressure relief valve actuation intensity to the sampling time. The pressure relief valve actuation intensity is used to describe the overall event strength corresponding to an out-of-phase segment, and it is used to provide the event contribution at the corresponding sampling time during the hourly calculation of the health degradation index. The operating strength of the pressure relief valve is determined by the following formula: ; In the formula, This indicates the intensity of the pressure relief valve action at the i-th sampling time. This represents the intensity of the pressure relief valve action corresponding to the k-th phase misalignment segment. This indicates that the i-th sampling time belongs to the k-th out-of-phase segment. The reason for using this mapping method is that the pressure relief valve action intensity is originally generated in segments, while the health degradation index is output along the time axis. Therefore, it is necessary to convert the segment-level action intensity into a sampling time-level state quantity. ; In the formula, This indicates the intensity of the pressure relief valve action at the i-th sampling time. This indicates that the i-th sampling moment does not belong to any out-of-phase segment, and a value of 0 indicates that no critical actuation intensity of the pressure relief valve was identified at that sampling moment. The reason for using this assignment method is that non-out-of-phase segment moments should not introduce a contribution of pressure relief valve actuation events to the health degradation index, thereby avoiding the incorrect evaluation of periods without events as an increased risk of valve actuation. Adaptive normalization based on median absolute deviation was performed on the compensated pressure sequence, oil temperature slow change, liquid level slow change, and pressure relief valve action intensity, respectively, to obtain the normalized results for compensated pressure, oil temperature, liquid level, and pressure relief valve action: ; In the formula, This represents the normalized result of the compensation pressure at the i-th sampling time. This represents the compensation pressure value at the i-th sampling time. This represents the sequence of compensating pressure values ​​within the current analysis period. This represents the median of the compensation pressure sequence. This represents the median absolute deviation of the compensating pressure series. This represents a very small positive number used to avoid a denominator of zero. The compensation pressure value needs to be converted into the degree of deviation relative to the current operating scale of the equipment in order to participate in the calculation of the health degradation index along with state quantities of different dimensions such as oil temperature, liquid level, and valve action intensity. ; In the formula, This represents the normalized oil temperature result at the i-th sampling time. This represents the slow variable of oil temperature at the i-th sampling time. This represents the sequence of slowly varying oil temperatures within the current analysis period. This represents the median of the oil temperature slowly varying series. This represents the absolute deviation of the median of the oil temperature variable series. This represents a very small positive number used to avoid a denominator of zero. Sustained high oil temperature typically reflects changes in thermal pressure and the heat dissipation state of the insulating oil. Normalizing the gradual change in oil temperature can generate a degradation component that can be integrated with the compensation pressure. ; In the formula, This represents the normalized liquid level result at the i-th sampling time. This represents the slow change in liquid level at the i-th sampling time. This represents the sequence of slowly changing liquid levels during the current analysis period. represents the median of the slow-moving liquid level series, and represents the absolute deviation of the median of the slow-moving liquid level series. This represents a very small positive number used to avoid a denominator of zero. Abnormal liquid level deviations may reflect changes in oil volume, gas chamber buffer status, or sensor measurement status. Normalizing the liquid level variable can serve as a stable input for health degradation assessment. ; In the formula, This represents the normalized result of the pressure relief valve action at the i-th sampling time. Let A represent the pressure relief valve actuation intensity at the i-th sampling time, and let A represent the sequence of pressure relief valve actuation intensities during the current analysis period. This represents the median of the sequence of pressure relief valve actuation intensity. This represents the median absolute deviation of the pressure relief valve actuation intensity sequence. This represents a very small positive number used to avoid a denominator of zero. The actuation intensity of the pressure relief valve is an event-type state quantity formed by phase-shifting segments. After normalization, it can be used together with the continuous compensation pressure, oil temperature, and liquid level state quantities in the degradation index calculation.

[0036] Since this embodiment focuses on deviations in the direction of health degradation, the data processing unit extracts the degradation component from each normalized result. The positive degradation component of the compensation pressure is calculated according to the following formula: ; In the formula, This represents the positive degradation component of the compensation pressure at the i-th sampling time. This represents the normalized result of the compensation pressure; a value of 0 indicates that the compensation pressure did not contribute positively to degradation. A compensation pressure higher than the current operating baseline of this equipment is a better indicator of pressure degradation risk, while a pressure deviation below the baseline should not be considered as positive degradation accumulation in this embodiment. The positive degradation component of oil temperature is calculated according to the following formula: ; In the formula, This represents the positive degradation component of oil temperature at the i-th sampling time. This indicates the normalized result of oil temperature. A value of 0 indicates that the oil temperature has not made a positive contribution to degradation. When the oil temperature is consistently higher than the operating reference, it is more likely to reflect the heat dissipation burden or thermal degradation. Oil temperatures lower than the reference are not considered as an accumulation of thermal degradation risk. The liquid level degradation component is calculated according to the following formula: ; In the formula, This represents the liquid level degradation component at the i-th sampling time. This represents the normalized liquid level result. The liquid level degradation component includes a high-level deviation component and a low-level deviation component. The high-level deviation component is determined by the portion of the normalized liquid level result that is greater than zero, and is used to characterize the contribution of oil expansion, changes in the gas chamber buffer state, or abnormal liquid level rise to health degradation. The low-level deviation component is determined by the absolute deviation portion of the normalized liquid level result that is less than zero, and is used to characterize the contribution of insufficient oil volume, oil level drop, or abnormal liquid level measurement to health degradation. The data processing unit takes the larger of the high-level and low-level deviation components as the liquid level degradation component and uses this component as the positive liquid level degradation contribution in the health degradation index. The reason for adopting this processing method is that liquid level anomalies in oil-immersed transformers can manifest as either high-level or low-level deviations, both of which affect oil insulation, heat dissipation, and the gas chamber buffer state. Therefore, bidirectional deviation conversion allows for a more complete contribution of the liquid level variation to the health degradation index, avoiding the omission of the risk of liquid level drop while only considering liquid level rise.

[0037] The positive degradation component of the pressure relief valve operation is calculated according to the following formula: ; In the formula, This represents the positive degradation component of the pressure relief valve action at the i-th sampling time. This represents the normalized result of the pressure relief valve's action. A value of 0 indicates that the pressure relief valve's action intensity did not contribute positively to the degradation. The higher the critical action intensity of the pressure relief valve, the more obvious the phase misalignment between the short-term pressure relief and the gradual change in liquid temperature. When used as an input for fault prediction and health management, this should be reflected as a positive risk contribution. The weights of each item are determined based on the relative dispersion of the compensated pressure sequence, oil temperature slow change, liquid level slow change, and pressure relief valve action intensity within a historical stable operating segment. Specifically, let the compensated pressure sequence, oil temperature slow change sequence, liquid level slow change sequence, and pressure relief valve action intensity sequence in the historical stable operating segment be respectively... , , and The weights of each item are calculated according to the following formula: ; In the formula, This indicates the weight corresponding to the compensating pressure. This represents the dispersion of compensating pressure within a historical stable operating segment. This represents the dispersion of the slowly varying oil temperature within a historical stable operating segment. This represents the dispersion of the slow-moving liquid level variation within a historical stable operating segment. This indicates the dispersion of the pressure relief valve's actuation intensity. This represents a very small positive number used to avoid a denominator of zero; the degree of dispersion of the compensation pressure in historical stable operating segments can characterize its relative contribution to state fluctuations, and determining the weights based on the degree of relative dispersion can avoid insufficient adaptability between different devices caused by artificially fixed weights. ; In the formula, This indicates the weight corresponding to the variable of oil temperature. This represents the dispersion of the slow-moving variable of oil temperature in the historical stable operating segment. The meanings of the other parameters are the same as those in the aforementioned weighting formula. The higher the relative dispersion of the slow-moving variable of oil temperature in the historical stable operating segment, the more its contribution to the fluctuation of normal state needs to be reasonably reflected in the health degradation index. ;

[0038] In the formula, This indicates the weight corresponding to the slow-moving variable of liquid level. This represents the dispersion of the slow-moving liquid level variable in a historical stable operating segment. The meanings of the other parameters are consistent with the aforementioned weighting formula. The stable fluctuation scale of the slow-moving liquid level variable can reflect the relative contribution of the liquid level state of the equipment to the overall operating state. Determining the weight based on this scale can make the health degradation index more consistent with the actual equipment state. ; In the formula, This indicates the weight corresponding to the intensity of the pressure relief valve's action. This represents the dispersion of the pressure relief valve's action intensity. The meanings of the other parameters are consistent with the aforementioned weighting formula. The pressure relief valve's action intensity is a key event information retained in this invention. Its weight should be determined based on the relative dispersion in historical stable operating segments to avoid misjudgment caused by differences in the sensitivity of pressure relief valves in different transformers. The dispersion of compensating pressure, the dispersion of oil temperature slow change, and the dispersion of liquid level slow change are determined according to the following formula: ; ; ; In the formula, This indicates the compensating pressure dispersion. This represents the compensation pressure sequence within a historically stable operating segment. This indicates the dispersion of the gradual change in oil temperature. This represents a sequence of slowly varying oil temperatures within a historical stable operating segment. This indicates the dispersion of the slow-moving liquid level. This represents a slow-moving sequence of liquid level values ​​within a historical stable operating segment. Indicates the absolute deviation of the median; It should be noted that the pressure relief valve actuation intensity is an event-type state variable. In historical stable operating segments, there may not be any critical actuation events for the pressure relief valve. In such cases, the pressure relief valve actuation intensity sequence in the historical stable operating segment may all be zero, and the corresponding median absolute deviation may also be zero. To avoid the pressure relief valve actuation intensity being excluded from the health degradation index calculation due to zero historical dispersion during actual phase-shifting segments, the data processing unit determines the dispersion corresponding to the pressure relief valve actuation intensity according to the following rules: If the median absolute deviation of the pressure relief valve actuation intensity sequence in the historical stable operating segment is not zero, then this median absolute deviation is used as the dispersion corresponding to the pressure relief valve actuation intensity; if the median absolute deviation of the pressure relief valve actuation intensity sequence in the historical stable operating segment is zero, and there is no non-zero pressure relief valve actuation intensity in the current analysis period, then the pressure relief valve actuation intensity does not participate in the health degradation index calculation in the current analysis period. Degradation index calculation: If the median absolute deviation of the pressure relief valve action intensity sequence in the historical stable operation segment is zero, but there is a non-zero pressure relief valve action intensity in the current analysis period, then the median absolute deviation of the non-zero pressure relief valve action intensity in the current analysis period is used as the dispersion corresponding to the pressure relief valve action intensity; if the number of non-zero pressure relief valve action intensities in the current analysis period is insufficient to calculate the median absolute deviation, then the median absolute value of the non-zero pressure relief valve action intensities in the current analysis period is used as the dispersion, and it is normalized together with the dispersion corresponding to the compensation pressure sequence, oil temperature slow variable and liquid level slow variable to determine the weight of each item.

[0039] After obtaining the various degradation components and their weights, the health degradation index is calculated. It should be noted that the health degradation index is a comprehensive state evaluation quantity formed by integrating the compensation pressure, oil temperature variation, liquid level variation, and pressure relief valve action intensity according to their corresponding weights. This index is used to represent the degree of deviation of the oil-immersed transformer from its operational health at the current sampling time, and the trend of its value reflects whether the equipment condition is developing in an unfavorable direction. The health degradation index is calculated using the following formula: ; In the formula, Let represent the health degradation index at the i-th sampling time. This indicates the weight corresponding to the compensating pressure. This indicates the positive degradation component of compensating pressure. This indicates the weight corresponding to the variable of oil temperature. This indicates the positive degradation component of oil temperature. This indicates the weight corresponding to the slow-moving variable of liquid level. This represents the liquid level degradation component, which is obtained by converting the high liquid level deviation component and the low liquid level deviation component. This indicates the weight corresponding to the intensity of the pressure relief valve's action. This represents the positive degradation component of the pressure relief valve's action. The reason for using this calculation method is that the compensated pressure reflects the continuous pressure state after equivalent compensation, the slow oil temperature and liquid level variables reflect the slow change state of the oil's thermal state and liquid level, and the pressure relief valve's action intensity reflects the critical action information of the valve retained by the compensation process. By weighting and integrating these four types of state variables, both data correction results and fault prediction information can be considered simultaneously, thus forming a comprehensive degradation evaluation oriented towards fault prediction and health management.

[0040] After obtaining the health degradation index, the fault prediction quantity is calculated based on the changing trend of the health degradation index. Specifically, the rate of change of the degradation index between the current sampling time and the previous sampling time is first calculated: ; In the formula, This represents the rate of change of the degradation index at the i-th sampling time. Let represent the health degradation index at the i-th sampling time. This represents the health degradation index at the (i-1)th sampling time. This represents the time interval between the current sampling time and the previous sampling time. The absolute value of the health degradation index reflects the current level of degradation, while the rate of change of the degradation index reflects whether the current degradation trend continues to intensify. The combination of the two can be used to assess short-term predictive risk.

[0041] The fault prediction value is calculated according to the following formula: ; In the formula, This represents the fault prediction value at the i-th sampling time. Let represent the health degradation index at the i-th sampling time. This represents the rate of change of the degradation index at the i-th sampling time. This indicates the prediction timescale. It's important to note that the predicted fault quantity is a trend assessment quantity extrapolated from the current health degradation index and its rate of change. It characterizes the level of health degradation that may be reached within the predicted timescale. The predicted fault quantity is not directly equivalent to a fault that has already occurred; rather, it is used to determine whether the current degradation trend has the potential to continue to intensify, and to provide a basis for background anomaly alerts and maintenance recommendations. The prediction timescale can be determined by the monitoring system's data upload cycle, the equipment operation and maintenance cycle, or the prediction cycle selected by the user. For example, it can be automatically configured based on the generation interval of each health report from the background system or the prediction duration that maintenance personnel are concerned about. The reason for using this calculation method is that when the current value of the health degradation index is already high and the rate of change is still positive, the predicted fault quantity will be higher than the current health degradation index, indicating that the degradation trend may continue to intensify within the predicted timescale. When the health degradation index tends to stabilize or decrease, the predicted fault quantity decreases accordingly, thus distinguishing between instantaneous high values ​​and continuous deterioration trends.

[0042] After the fault prediction quantity is generated, the data processing unit further generates health management results that can be used for background display, abnormal alarms, and maintenance suggestions. Specifically, it outputs the compensated pressure sequence, oil temperature slow change sequence, liquid level slow change sequence, phase misalignment segment, pressure relief valve action intensity, health degradation index, and fault prediction quantity. The output results are expressed as follows: ; In the formula, This represents the set of health management outputs at the i-th sampling time and its associated phase-out segments. Indicates the compensation pressure value. Indicates a gradual change in oil temperature. Indicates a slow change in liquid level. This represents the k-th phase misalignment segment. This represents the intensity of the pressure relief valve action corresponding to the k-th phase misalignment segment. Let represent the health degradation index at the i-th sampling time. This represents the fault prediction value at the i-th sampling time. The compensated pressure value is used to replace the original pressure value in the pressure status display and subsequent status judgment. The oil temperature gradual variable and liquid level gradual variable are used to display the gradual change state of temperature and liquid. The phase misalignment segment and pressure relief valve action intensity are used to indicate the critical action of the pressure relief valve. The health degradation index is used to represent the current degree of health degradation. The fault prediction value is used to represent the degradation risk within the predicted time scale.

[0043] For critical operation alerts of pressure relief valves, the data processing unit generates event alert content based on phase misalignment segments and the intensity of pressure relief valve operation. When a phase misalignment segment exists and the corresponding pressure relief valve operation intensity deviates significantly positively from historical stable operation segments, the system marks the phase misalignment segment as a critical operation alert for the pressure relief valve and marks the start time, end time, pressure step drop, and pressure relief valve operation intensity of the segment in the data curve. This allows maintenance personnel to confirm that the short-term pressure drop is not a typical pressure sensor glitch and should not be simply interpreted as a true recovery of internal pressure. For transformer health status alerts, the data processing unit generates status output based on the compensated pressure sequence, health degradation index, and fault prediction quantity. When the compensated pressure sequence is continuously higher than its own operating baseline, the health degradation index continues to increase, or the fault prediction quantity is higher than the health degradation index and continues to increase for multiple consecutive sampling periods, the system generates a transformer health status alert, indicating that the current pressure thermal state and valve operation events continuously contribute to health degradation. For maintenance recommendations, the data processing unit generates maintenance recommendations when the fault prediction quantity continues to increase relative to the health degradation index. The maintenance recommendations may include recommending the review of pressure relief valve action records, checking the installation and calibration status of pressure sensors, verifying the continuity of oil temperature and liquid level sensor data, checking the buffer status of oil tank or gas chamber, and arranging subsequent inspections. However, the specific text of the maintenance recommendations can be configured by the user management module or the equipment management module according to the operation and maintenance rules.

[0044] Furthermore, the continuous increase of the fault prediction quantity relative to the health degradation index means that within multiple consecutive valid sampling periods, the fault prediction quantity is higher than the health degradation index at the corresponding sampling time, and the difference between the fault prediction quantity and the health degradation index does not show a continuous decline. These multiple consecutive valid sampling periods can be configured by the background maintenance strategy or determined by the number of valid sampling points within a single health report period. When there are data gaps, communication anomalies, or sensor over-range anomalies within a consecutive valid sampling period, the abnormal sampling period is not included in the persistence judgment. If the fault prediction quantity continuously increases relative to the health degradation index, and this persistent state covers the multiple consecutive valid sampling periods, the data processing unit generates a maintenance suggestion or increases the maintenance suggestion level. If the fault prediction quantity decreases relative to the health degradation index, tends to stabilize, or only increases at isolated sampling times, the data processing unit retains the status record but does not increase the maintenance suggestion level. Through this method, maintenance suggestions no longer rely solely on the instantaneous results of a single sampling time, but are generated based on the continuous changing trend of the fault prediction quantity, thereby reducing the risk of false alarms caused by instantaneous disturbances.

[0045] Regarding abnormal data processing, if a pressure, oil temperature, or liquid level value is missing at a certain sampling moment, the data processing unit first processes it based on the duration of the missing value. When the missing value only occurs in a few isolated sampling points, the same sensor can be used to perform linear interpolation on adjacent valid sampling points or maintain the previous valid value to form a temporary calculated value, and the sampling moment is marked as the interpolation state. When the duration of the missing value exceeds the allowable range of the data window, the data processing unit stops extracting out-of-phase segments for that period and outputs a data insufficiency prompt to avoid misidentification of pressure relief valve action due to missing data. If the pressure sensor shows an over-range value, the oil temperature sensor shows a sudden jump value that does not conform to the continuity of physical changes, or the liquid level sensor shows a communication anomaly value, the data processing unit marks the anomaly value before entering the median smoothing and pressure residual calculation, and will not directly use the marked anomaly value for fitting historical stable operation segments. For anomalies that cannot be removed in the current online calculation, median smoothing and median absolute deviation normalization can reduce the impact of single-point anomalies on the gradual temperature and liquid change state, out-of-phase intensity, and health degradation index.

[0046] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for intelligent compensation of multi-sensor data in oil-immersed transformers, characterized in that, Includes the following steps: S1. Obtain the pressure sequence, oil temperature sequence, and liquid level sequence of the oil-immersed transformer at the same time axis; S2. Determine the slowly changing temperature and liquid state based on the oil temperature sequence and liquid level sequence, and construct a slowly changing pressure reference quantity based on the slowly changing temperature and liquid state. Obtain the pressure residual based on the deviation relationship between the pressure sequence and the slowly changing pressure reference quantity. S3. Based on the rapid decline of the pressure residual and the gradual change of the temperature liquid, determine the phase misalignment state between the rapid decline of pressure and the gradual change of temperature liquid, and extract the corresponding phase misalignment segment from the time axis. S4. Determine the pressure step fall characterization quantity based on the pressure residual change in the phase-misaligned segment, and perform equivalent backfilling on the pressure sequence based on the pressure step fall characterization quantity to generate a compensated pressure sequence. Determine the pressure relief valve action intensity quantity based on the pressure step fall characterization quantity and the phase misalignment degree of the phase-misaligned segment. S5. Calculate the health degradation index of the oil-immersed transformer based on the compensation pressure sequence, the gradual change state of the liquid temperature, and the action intensity of the pressure relief valve, and calculate the fault prediction quantity based on the changing trend of the health degradation index.

2. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 1, characterized in that, The slowly changing temperature-liquid state is determined based on the oil temperature and liquid level sequences, and a slowly changing pressure reference quantity is constructed based on this state. The pressure residual is obtained based on the deviation between the pressure sequence and the slowly changing pressure reference quantity, including: A data window is constructed based on data adjacent to the current sampling time; The median of the oil temperature data and the liquid level data within the data window are calculated respectively, and the median is used as the oil temperature slow variable and the liquid level slow variable at the corresponding sampling time. The oil temperature slow variable and the liquid level slow variable together constitute the temperature and liquid slow change state. Obtain the pressure baseline coefficient, the first pressure influence coefficient, and the second pressure influence coefficient obtained by fitting historical stable operation segments of the transformer; Based on the pressure reference coefficient, the first pressure influence coefficient, the second pressure influence coefficient, the oil temperature slow change variable, and the liquid level slow change variable, a pressure slow change reference quantity is obtained; The pressure residual is obtained by subtracting the pressure sequence at the same sampling time from the slowly varying pressure reference.

3. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 2, characterized in that, Based on the rapid decline of the pressure residual and the gradual change of the temperature liquid, the phase misalignment state between the rapid pressure decline and the gradual temperature change is determined, and the corresponding phase misalignment segments are extracted from the time axis, including: Calculate the change in pressure residual, the gradual change in oil temperature, and the gradual change in liquid level between the current time and the previous time. Extract the positive part of the negative value of the pressure residual change, and extract the absolute value of the gradual change in oil temperature and the absolute value of the gradual change in liquid level. Adaptive normalization is performed on the positive part of the negative value of the pressure residual change, the absolute value of the oil temperature change, and the absolute value of the liquid level change by means of the median absolute deviation, thereby obtaining the pressure step drop intensity, the oil temperature change intensity, and the liquid level change intensity in sequence. Calculate the sum of the intensity of the gradual change in oil temperature and the intensity of the gradual change in liquid level, add the sum to the value of 1, and take the reciprocal to obtain the gradual change inhibition factor; Multiplying the pressure step fall intensity by the gradual suppression factor yields the phase misalignment intensity; The continuous time period during which the phase fault intensity exhibits a local peak and the pressure residual changes from decreasing to stabilizing or increasing is defined as a phase fault segment.

4. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 2, characterized in that, The pressure step fall characterization quantity is determined based on the pressure residual change within the phase-shifted segment, and the pressure sequence is equivalently compensated based on the pressure step fall characterization quantity to generate a compensated pressure sequence, including: The difference between the pressure residual at the beginning of the phase-shifted segment and the minimum pressure residual within the phase-shifted segment is taken as the pressure step drop. The pressure recovery weight is obtained by dividing the difference between the pressure residual at the beginning of the phase misalignment segment and the pressure residual at the current moment by the pressure step drop amount. The pressure recovery weight is limited to a range that is not less than zero and not greater than one. For the sampling time belonging to the phase misalignment segment, the product of the pressure recovery weight and the pressure step drop amount at the corresponding time is added to the original pressure value to obtain the compensation pressure value within the segment. For sampling times that do not belong to the phase-shifted segment, the original pressure value is used as the external compensation pressure value, and the internal compensation pressure value and the external compensation pressure value together constitute the compensation pressure sequence.

5. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 3, characterized in that, The pressure relief valve's actuation intensity is determined based on the pressure step fall characteristic and the degree of phase misalignment of the misaligned segment, including: Calculate the arithmetic mean of the phase fault intensity at all sampling times within the phase fault segment; Multiply the pressure step drop by the arithmetic mean of the phase misalignment intensity to obtain the pressure relief valve action intensity corresponding to the phase misalignment segment.

6. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 2, characterized in that, Based on the compensation pressure sequence, the gradual change state of the liquid temperature, and the intensity of the pressure relief valve action, the health degradation index of the oil-immersed transformer is calculated, including: The pressure relief valve actuation intensity is mapped to the sampling time of the corresponding phase-out segment, and the value is set to zero at the non-phase-out segment time to obtain the pressure relief valve actuation intensity distributed on the time axis; Adaptive normalization based on median absolute deviation is performed on the compensation pressure sequence, the oil temperature variable, the liquid level variable, and the pressure relief valve action intensity, respectively, to obtain the compensation pressure normalization result, oil temperature normalization result, liquid level normalization result, and pressure relief valve action normalization result. Extract the portions greater than zero from the normalized results of the compensation pressure, the normalized results of the oil temperature, the normalized results of the liquid level, and the normalized results of the pressure relief valve action, respectively, as the corresponding positive degradation components for each item; The weights of each item are determined based on the relative dispersion of the compensation pressure sequence, the oil temperature variable, the liquid level variable, and the pressure relief valve action intensity in historical stable operating segments. The health degradation index is obtained by weighting and summing the positive degradation components of each item with their corresponding weights.

7. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 6, characterized in that, Based on the relative dispersion of the compensated pressure sequence, the oil temperature buffer, the liquid level buffer, and the pressure relief valve action intensity in historical stable operating segments, the weights corresponding to each item are determined, including: Calculate the median absolute deviation of the compensation pressure sequence, the oil temperature slow variable, the liquid level slow variable, and the pressure relief valve action intensity in the historical stable operation segment; and use the ratio of the median absolute deviation of each item to the sum of the median absolute deviations of all items as the weight corresponding to that item.

8. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 1, characterized in that, Failure prediction is calculated based on the changing trend of the health degradation index, including: Calculate the difference between the health degradation index at the current time and the previous time, and divide the difference by the time interval between the current time and the previous time to obtain the rate of change of the degradation index; The rate of change of the degradation index is multiplied by the prediction time scale, and the product is added to the health degradation index at the current time to obtain the fault prediction amount. The prediction timescale is determined by the monitoring system data upload cycle, the equipment operation and maintenance cycle, or the prediction cycle selected by the user.

9. The intelligent compensation method for multi-sensor data of an oil-immersed transformer according to claim 1, characterized in that, After obtaining the fault prediction value, it also includes: Generate a critical action indication for the pressure relief valve based on the phase misalignment segment and the actuation intensity of the pressure relief valve; Transformer health status indicators are generated based on compensation pressure sequence, health degradation index, and fault prediction. Maintenance recommendations are generated when the number of predicted failures continues to increase relative to the health degradation index.

Images