Transformer winding temperature rise on-line monitoring method and system based on multi-point temperature sensing

Abstract

The present application relates to the technical field of power equipment state monitoring, and more particularly to a transformer winding temperature rise online monitoring method and system based on multi-point temperature sensing. Multi-point temperature values and operation quantity data of the transformer are collected; a thermal event time window is established to align multi-source data; temperature prediction values and hotspot temperature prior values are calculated based on a winding lumped parameter thermal circuit model; residuals are calculated and the health status and sensitivity of the measuring points are evaluated; the importance of the measuring points is split into observation importance negatively correlated with the residuals and review importance positively correlated with the residuals; sampling scheduling is adjusted accordingly; candidate models are verified in parallel before replacing the stable model; cooling effect is evaluated and feedback correction is performed; the observation importance of abnormal measuring points is weakened, degradation estimation is performed and an alarm is output. The present application provides a transformer winding temperature rise online monitoring method and system based on multi-point temperature sensing, reduces the pollution of abnormal measuring points on hotspot estimation, and realizes quantitative evaluation and closed-loop correction of cooling effect.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and in particular to a method and system for online monitoring of transformer winding temperature rise based on multi-point temperature sensing. Background Technology

[0002] The hot spot temperature of a transformer winding is a key parameter determining the insulation life and load capacity of an oil-immersed power transformer. Existing online monitoring technologies typically collect temperature values ​​at different locations on the winding, as well as operating parameters such as load current, oil temperature, and cooling equipment status, and estimate the hot spot temperature of the winding using thermal circuit models or empirical formulas.

[0003] In engineering practice, the above monitoring methods have the following shortcomings. First, data such as temperature, load current, and cooling status come from different acquisition devices, each with different sampling periods and communication delays. If data is processed directly according to its arrival order, situations may arise where the load has changed but the temperature has not yet responded, or cooling has been activated but the windings have not cooled down due to thermal inertia. This results in a lack of a unified time reference for temperature rise calculation and cooling intervention. Second, the contribution of different temperature measurement points to the estimation of winding hot spot temperature varies significantly. Measurement points closer to the hot spot area are more sensitive to temperature rise judgment, while those farther away contribute less. However, existing systems typically use the same sampling frequency for all measurement points, making it impossible to allocate sampling resources to high-value measurement points during critical events. Third, individual temperature sensors may drift, become intermittent, or experience communication anomalies. The conventional approach is to directly remove abnormal measurement points or merge them according to fixed weights. This lacks refined management of abnormal measurement points, failing to effectively isolate abnormal data from contaminating hot spot estimation and failing to maintain continuous monitoring of abnormal measurement points so that they can be promptly reintegrated when they return to normal. Fourth, transformer thermal model parameters may shift with increasing operating years and changes in oil flow characteristics. If model parameters are updated directly using data from a single operation, deviations caused by abnormal measurement points or transient operating conditions are easily written into the model incorrectly, leading to amplified alarm deviations in subsequent alarms. Fifth, existing systems mostly determine whether cooling action is complete based on cooling equipment status signals. However, insufficient fan output, decreased oil pump efficiency, or continued load increases can all cause hot spot temperatures to continue rising after cooling equipment is activated. The system lacks evaluation of the actual cooling temperature rise suppression effect and means to correct subsequent prediction and control strategies.

[0004] The aforementioned shortcomings lead to problems in existing online monitoring solutions during engineering applications, such as large deviations in hotspot temperature estimation, long detection times for abnormal measuring points, and a lack of evaluation of cooling effects. Specifically, inconsistent time bases cause a time misalignment error between the calculated temperature rise and the actual thermal state. This error is particularly pronounced during periods of rapid load or cooling changes, potentially causing hotspot temperature estimation deviations of several degrees Celsius. If abnormal measuring points are included in the fusion process with conventional fixed weights, individual sensor malfunctions can cause hotspot temperature estimates to deviate from the true values, resulting in false alarms or missed alarms. The lack of effectiveness evaluation after cooling equipment is put into operation prevents the system from distinguishing between normal cooling response and deteriorated cooling capacity, potentially delaying maintenance of cooling equipment or causing improper over-temperature protection activation.

[0005] Therefore, a transformer winding temperature rise online monitoring scheme is needed that integrates thermal event time window alignment, measurement point importance function decomposition, thermal state prediction estimation, parallel model version verification, and cooling effect feedback correction. This scheme should correspond the hot spot temperature estimation output with the measurement point health status, sampling scheduling strategy, and model credibility, and output degradation estimates and alarm information under abnormal measurement point degradation conditions. Summary of the Invention

[0006] To achieve the above-mentioned objectives, this invention provides a method and system for online monitoring of transformer winding temperature rise based on multi-point temperature sensing. This system addresses issues such as hot spot temperature estimation deviations caused by abnormal measurement point data contamination, time misalignments in thermal state calculations caused by asynchronous multi-source data sampling periods and communication delays, and the inability to detect cooling capacity degradation due to a lack of quantitative evaluation of the effects of cooling equipment after its implementation.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for online monitoring of transformer winding temperature rise based on multi-point temperature sensing, comprising:

[0008] Collect multiple temperature values ​​and operating data of the transformer;

[0009] Establish hot event time windows to align multi-source data;

[0010] The predicted temperature and the prior hot spot temperature are calculated based on the lumped parameter thermal circuit model of the winding.

[0011] Calculate the residuals and assess the health status and sensitivity of the measuring points;

[0012] The importance of measurement points is divided into observation importance and verification importance. The observation importance is negatively correlated with the residuals, and the verification importance is positively correlated with the residuals.

[0013] The sampling schedule is adjusted based on the observation importance and the verification importance.

[0014] Candidate models can only replace stable models after being validated in parallel.

[0015] Evaluate the cooling effect after the cooling system is put into operation and provide feedback for correction.

[0016] After reducing the observation importance of abnormal measurement points, a downgraded estimate is performed and an alarm is output.

[0017] To further realize the present invention, the following technical solutions may be preferred:

[0018] Preferably, establishing a thermal event time window to align multi-source data includes:

[0019] When the load current change exceeds the load threshold, or the cooling equipment status changes, or the temperature rise rate of the measuring point continuously exceeds the temperature rise rate threshold, the earliest time that meets the conditions is taken as the starting point of the thermal event, and the first preset time is extended forward and the second preset time is extended backward to form a thermal event time window.

[0020] When the rate of change of temperature at a measuring point exceeds the thermal inertia constraint boundary and there is no load, oil temperature or cooling change attributable to it, the measuring point is marked as abnormal and the measuring point abnormality handling is triggered, rather than being used as a trigger condition for establishing a thermal event time window.

[0021] Within the thermal event time window, the switch data is processed by holding the most recently valid state, and the continuous data is processed by linear interpolation or holding the most recently valid data. Data with severe missing data or time stamp exceeding the limit is marked as invalid.

[0022] Preferably, the calculation of the predicted temperature value and the prior value of the hot spot temperature based on the winding lumped parameter thermal circuit model includes:

[0023] A lumped-parameter thermal circuit model for windings, established based on thermoelectric analogy, is used. This model includes nonlinear thermal resistance, thermal capacity, and thermal conductivity reflecting oil flow and cooling effects. The initial model parameters of the lumped-parameter thermal circuit model are determined based on transformer structural parameters, factory test data, type test data, and / or historical operating data. The state transition equation of the lumped-parameter thermal circuit model is recursively derived from the posterior estimate of the thermal state of the previous time window, the square of the current load current, oil temperature, ambient temperature, and the state of the cooling equipment corrected by the cooling efficiency coefficient. The predicted temperature values ​​at each measuring point are calculated from the predicted thermal state values, and the prior values ​​of the winding hot spot temperatures are extracted.

[0024] Its state transition equation is as follows:

[0025]

[0026] in, It is a state vector containing the winding hot spot temperature and the equivalent temperature of the corresponding node at each measuring point. The state transition matrix is ​​determined by the thermal time constant. For the input matrix, The input vector includes the square of the load current, the top oil temperature, the ambient temperature, and the cooling state corrected by the cooling efficiency coefficient. Based on the posterior estimate of the thermal state of the previous time window, the square of the current load current, the oil temperature, the ambient temperature, and the cooling equipment status, the predicted temperature values ​​of each measuring point and the prior values ​​of the winding hot spot temperature of the current time window are recursively derived.

[0027] Preferably, the step of splitting the importance of measurement points into observation importance and verification importance, wherein the observation importance is negatively correlated with the residuals and the verification importance is positively correlated with the residuals, includes:

[0028] The residual amplitude and sensitivity are normalized respectively to obtain the normalized residual and normalized sensitivity;

[0029] The observation importance is determined by the product of the health status factor, the normalized sensitivity, and 1 minus the normalized residual, so that the health-sensitive measurement point with the smaller residual has a higher observation importance, which is used to determine the participation weight of the measurement point in the hotspot temperature fusion estimation.

[0030] The review importance is determined by the sum of the first component and the second component, wherein the first component is the product of the health status factor, the normalized sensitivity and the normalized residual, and the second component is the product of the preset compensation coefficient, the difference between 1 and the health status factor and the normalized sensitivity, so that the review importance obtained by sensitive measurement points with abnormal residuals or key measurement points of health deterioration is higher, and is used to determine the sampling review frequency of the measurement point.

[0031] Specifically, the importance of the observation is determined by the formula... Confirmed, among which For measuring points Health status factors For normalization sensitivity, To normalize the residuals, health-sensitive monitoring points with smaller residuals are assigned higher observation importance, which is used to determine the participation weight of the monitoring point in hotspot temperature fusion estimation; the verification importance is calculated using the formula... Confirmed, among which The preset compensation coefficient ranges from 0.3 to 0.8.

[0032] Preferably, adjusting the sampling schedule based on the observation importance and the verification importance includes:

[0033] The sampling interval for each measuring point in the next time window is determined by the quotient obtained by dividing the baseline sampling interval by 1 and summing the sum of the linear combination of the observation importance and the verification importance, and the quotient is limited to the range between the minimum sensor response interval and the maximum monitoring requirement interval; specifically, the sampling interval for each measuring point in the next time window is determined by the formula... Determine and limit the resulting quotient to between the minimum sensor response interval and the maximum monitoring requirement interval, where As the reference sampling interval, and These are positive weighting coefficients;

[0034] When the sensor does not support independent adjustment of the sampling frequency by measurement point, the sampling schedule is adjusted to increase the frequency of high-importance measurement points in the polling sequence. Through the master station's time-sharing scheduling and priority queue mechanism, conflict-free non-equal interval polling is achieved under the bus communication protocol.

[0035] Preferably, the candidate model can only replace the stable model after parallel verification, including:

[0036] When the data involved in the update is complete and valid, the health status factors of the measurement points involved in the update are all higher than the minimum confidence threshold, the number of measurement points involved in the update is not less than the preset lower limit and the spatial distribution covers the main area of ​​the winding, and the minimum singular value of the matrix synthesized by the weight matrix constructed from the importance of measurement point observation and health factors and the model output sensitivity matrix to parameters is not lower than the information content threshold, the model parameter update is deemed to be valid.

[0037] With the goal of reducing the weighted sum of squared residuals, candidate model parameters are generated using the sensitivity matrix and regularization term. The candidate model parameters are then restricted to a physically feasible range determined by the thermal time constant and thermal resistance and heat capacity. If the parameters exceed the boundary of the feasible range, they are adjusted to the boundary values.

[0038] The candidate model and the stable model are run in parallel over multiple consecutive time windows. When the observation importance weighted residual of the candidate model is consistently lower than that of the stable model and the hot spot estimate and alarm level do not fluctuate abnormally, the candidate model is switched to a new stable model. Otherwise, the candidate model is discarded and the original stable model is maintained.

[0039] Preferably, the evaluation of the cooling effect and feedback for correction after the cooling is implemented includes:

[0040] When the cooling equipment is switched from being deactivated to being activated, a cooling feedback window is established that includes thermal inertia delay time and effect evaluation time.

[0041] Calculate the improvement in temperature rise rate, load square change, and oil temperature change before and after the cooling action. Subtract the load square change correction term and oil temperature change correction term from the improvement in temperature rise rate to obtain the cooling effect evaluation quantity.

[0042] When the cooling effect evaluation value is lower than the effective threshold, the cooling efficiency coefficient is lowered by a preset step size or the cooling recommendation trigger temperature threshold is reduced.

[0043] Preferably, the step of reducing the observation importance of abnormal measurement points, performing downgrading estimation, and outputting alarms includes:

[0044] When the health status factor of a measuring point is lower than the abnormal threshold, the observation importance of the measuring point is automatically made close to zero through the calculation method of the observation importance, and it does not directly participate in the hot spot temperature fusion estimation.

[0045] The degradation estimate is determined by a weighted combination of the thermal model prediction, the temperature value of adjacent healthy monitoring points weighted by spatial distance, the oil temperature and the historical trend extrapolation value, and is accompanied by a degradation monitoring status indicator.

[0046] When the health status factor of the abnormal measurement point returns to normal, the residual returns to the normal range, and the spatial consistency is restored within multiple consecutive time windows, its normal participation status is gradually restored.

[0047] A transformer winding temperature rise online monitoring system based on multi-point temperature sensing includes a multi-point temperature acquisition unit, an operation quantity acquisition unit, a thermal event time window management unit, a thermal state prediction unit, a measuring point state evaluation unit, a measuring point importance calculation unit, a sampling scheduling unit, a model version management unit, a cooling feedback evaluation and correction unit, and a degradation estimation and alarm output unit.

[0048] The output terminals of the multi-point temperature acquisition unit and the operation data acquisition unit are connected to the input terminal of the thermal event time window management unit, and are used to transmit the acquired temperature values ​​and operation data to the thermal event time window management unit for time calculation.

[0049] The output of the thermal event time window management unit is connected to the input of the thermal state prediction unit, and is used to provide it with time-aligned data.

[0050] The output terminals of the thermal state prediction unit and the thermal event time window management unit are respectively connected to the input terminal of the measuring point state evaluation unit. The measuring point state evaluation unit is used to calculate residuals, evaluate health status factors, and determine sensitivity.

[0051] The output of the measuring point status evaluation unit is connected to the input of the measuring point importance calculation unit, which is used to generate observation importance and verification importance.

[0052] The output of the measurement point importance calculation unit is connected to the input of the sampling scheduling unit and the degradation estimation and alarm output unit, respectively. The sampling scheduling unit is used to adjust the sampling scheduling, and the degradation estimation and alarm output unit is used to correct the hot spot temperature and output an alarm.

[0053] The model version management unit is bidirectionally connected to the thermal state prediction unit, and is used to load model parameters and perform parallel verification.

[0054] The cooling feedback evaluation and correction unit is bidirectionally connected to the thermal state prediction unit and is used to evaluate the cooling effect and correct the cooling parameters.

[0055] Preferably, the model version management unit is also used to perform parallel operations on the candidate model and the stable model in multiple consecutive time windows. When the observation importance weighted residual of the candidate model is continuously lower than that of the stable model and the hot spot estimate and alarm level do not fluctuate abnormally, the candidate model is switched to a new stable model. The alarm output is based solely on the hot spot temperature estimate output by the stable model version.

[0056] The beneficial effects of this invention are:

[0057] This invention decomposes the importance of measurement points into observation importance and verification importance, giving higher weight to healthy and sensitive measurement points with small residuals in hotspot estimation, and increasing the sampling and verification frequency of sensitive measurement points with abnormal residuals. This automatically isolates abnormal measurement points during the estimation stage and maintains continuous tracking of abnormal measurement points during the monitoring stage, reducing the contamination of hotspot estimation by abnormal measurement points and shortening the anomaly detection time. By establishing a unified time benchmark for multi-source data through a thermal event time window, it reduces time misalignment errors caused by different sampling periods and communication delays, improving the accuracy of temperature rise calculations during load changes and cooling actions. Candidate models are replaced by stable models only after parallel verification, and alarm outputs are always based on the stable model version, avoiding interference from abnormal data to the alarm system during online parameter updates. A cooling feedback evaluation and correction mechanism enables quantitative evaluation of cooling effects and closed-loop correction of cooling parameters, allowing the cooling efficiency coefficient and cooling intervention conditions to adaptively adjust with changes in actual cooling capacity. Attached Figure Description

[0058] Figure 1 This is a topology diagram of the overall system architecture of the present invention.

[0059] Figure 2 This is a schematic diagram illustrating the principle of thermal event time window construction and data arithmetic in this invention.

[0060] Figure 3 This is a schematic diagram of the measurement point importance decomposition and sampling scheduling process of the present invention.

[0061] Figure 4 This is a schematic diagram illustrating the parallel verification and switching principle of the model version management unit of the present invention.

[0062] Figure 5 This is a schematic diagram of the cooling feedback evaluation and correction closed-loop principle of the present invention. Detailed Implementation

[0063] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Example 1

[0066] See Figure 1 As shown, in one specific embodiment, the online monitoring system of the present invention is deployed on an oil-immersed power transformer. The system includes a multi-point temperature acquisition unit, an operation quantity acquisition unit, a thermal event time window management unit, a thermal state prediction unit, a measuring point state evaluation unit, a measuring point importance calculation unit, a sampling scheduling unit, a model version management unit, a cooling feedback evaluation and correction unit, and a degradation estimation and alarm output unit. The output terminals of the multi-point temperature acquisition unit and the operation quantity acquisition unit are connected to the input terminal of the thermal event time window management unit, used to transmit the acquired temperature values ​​and operation quantity data to the thermal event time window management unit. The output terminal of the thermal event time window management unit is connected to the input terminal of the thermal state prediction unit, used to provide it with time-aligned operation quantity and cooling state data. The output terminal of the thermal state prediction unit and the temperature output terminal of the thermal event time window management unit are respectively connected to the input terminal of the measuring point state evaluation unit. The output terminal of the measuring point state evaluation unit is connected to the input terminal of the measuring point importance calculation unit. The first output of the measurement point importance calculation unit is connected to the input of the sampling scheduling unit, and the second output is connected to the input of the model version management unit. The model version management unit is bidirectionally connected to the thermal state prediction unit. The cooling state output of the thermal event time window management unit is connected to the input of the cooling feedback evaluation and correction unit, which is bidirectionally connected to the thermal state prediction unit. The outputs of the measurement point importance calculation unit and the thermal state prediction unit are also connected to the input of the degradation estimation and alarm output unit.

[0067] In this embodiment, the multi-point temperature acquisition unit includes temperature sensors deployed at multiple locations on the transformer. For non-winding internal measuring points such as the temperature-measuring blind tube on the transformer tank wall, near clamps, temperature-measuring bushings, top oil temperature location, bottom oil temperature location, and ambient location, PT100 platinum resistance temperature sensors can be selected. When it is necessary to directly measure the temperature inside the windings or in the vicinity of hot spots, fiber optic grating temperature sensors are used at the corresponding locations. The number of temperature sensors can be set to, for example, eight, but can be increased or decreased depending on the transformer capacity and winding structure. Each temperature sensor is connected to the acquisition module via cable or optical fiber, and the acquisition module has analog-to-digital conversion capabilities and a communication interface.

[0068] The operational data acquisition unit is used to acquire three-phase current signals from the secondary side of the existing current transformers in the transformer, the top oil temperature value from the oil surface thermometer, and the status signals of the cooling equipment from the cooling control cabinet. The operational data acquisition unit is also used to acquire ambient temperature data from the substation environmental monitoring station.

[0069] The thermal event time window management unit, thermal state prediction unit, measuring point state evaluation unit, measuring point importance calculation unit, sampling scheduling unit, model version management unit, and cooling feedback evaluation and correction unit can be integrated into an embedded processing device. This embedded processing device is deployed within the transformer's local control cabinet or in the edge computing node of the substation's secondary equipment room. The degradation estimation and alarm output unit can be deployed integrally with the embedded processing device or independently on the station control layer's backend server.

[0070] Example 2

[0071] In this embodiment, the online monitoring method is executed cyclically according to the following steps.

[0072] Step S1: Collect multiple temperature values ​​and operating data of the transformer.

[0073] Specifically, the multi-point temperature acquisition unit reads temperature values ​​from each temperature measurement point according to the currently effective sampling scheduling instruction, at their respective sampling intervals, and denoted as... ,in Number the measurement points. This is the current time window number. The multi-point temperature acquisition unit simultaneously records the sampling time and data quality identifier for each temperature value.

[0074] The operation data acquisition unit obtains the equivalent load current within the current time window. Top oil temperature Ambient temperature and cooling equipment status The equivalent load current can be calculated from the root mean square value of the three-phase current, or the maximum or average value of the three-phase current can be taken. The cooling equipment status includes at least the fan on / off status and the oil pump on / off status. When grouped cooling status is available, the cooling equipment status... Expand to vector form.

[0075] Furthermore, the operational data acquisition unit adds a time stamp to all acquired data. The accuracy of the time stamp is aligned with the sampling time of the multi-point temperature acquisition unit to the same time reference, which can be taken from the substation's GPS or BeiDou time synchronization system.

[0076] Step S2: Establish hot event time windows to align multi-source data.

[0077] See Figure 2 As shown, the thermal event time window management unit continuously monitors the changes in equivalent load current, changes in cooling equipment status, the rate of temperature rise at each measuring point, and the rate of temperature change at each measuring point, and identifies the starting point of the thermal event according to the following rules.

[0078] A load change event is triggered when the change in equivalent load current exceeds a load change threshold within a set time period. The set time period can be, for example, 10 to 60 seconds, and the load change threshold can be, for example, 5% to 15% of the rated current, or adjusted according to the specific capacity and operating characteristics of the transformer.

[0079] A cooling action event is triggered when the state of the cooling equipment changes.

[0080] A temperature rise event is triggered when the rate of temperature change at a certain measuring point exceeds a temperature rise rate threshold for multiple consecutive sampling cycles. The temperature rise rate threshold can be set to, for example, from 0.5℃ / min to 2.0℃ / min.

[0081] When multiple event conditions are met simultaneously, the earliest time point when the conditions are met is taken as the starting point of the hot event.

[0082] When the rate of temperature change at a certain measuring point exceeds the theoretical upper limit determined by the transformer's thermal time constant, and it is determined that this rate of change cannot be explained by load changes, oil temperature changes, or cooling state changes, this phenomenon is usually caused by sensor hardware failure or communication interference. The system does not treat this as the starting point of a global thermal event, but directly marks the measuring point as abnormal and proceeds to step S9 for measuring point abnormality processing. This theoretical upper limit can be set as 3℃ / min to 8℃ / min.

[0083] The hotspot temperature rise rate obtained from the previous calculation cycle or rolling estimate can also be used to trigger a temperature rise change event when it exceeds the temperature rise rate threshold, thus avoiding the problem that it cannot be triggered because the hotspot temperature estimate has not been completed in the current cycle.

[0084] Furthermore, based on the starting point of the thermal event, a first preset duration is extended forward and a second preset duration is extended backward to form a thermal event time window. The first preset duration can be set to 3 to 10 minutes for example, and the second preset duration can be set to 10 to 30 minutes for example. Both can be adjusted according to the transformer thermal time constant and the load change rate.

[0085] For routine monitoring periods when no thermal events occur, the thermal event time window management unit operates using a rolling time window method.

[0086] Within the thermal event time window, the thermal event time window management unit aggregates temperature values ​​and operational data from different sampling periods into a unified discrete calculation time sequence. For switch-type data such as cooling status, the most recent valid state is retained. For continuous data such as oil temperature and ambient temperature, linear interpolation is used; when data is missing, the most recent valid data is retained. Data with severe missing data or timescale deviations exceeding the allowable range is marked as invalid in the data quality identifier.

[0087] Step S3: Calculate the predicted temperature value and the prior value of the hot spot temperature based on the winding lumped parameter thermal circuit model.

[0088] Specifically, the thermal state prediction unit employs a winding lumped parameter thermal circuit model based on the thermoelectric analogy principle. This model equates the transformer's heat conduction, convection, and radiation processes to a thermal conduction network that includes nonlinear thermal resistance, thermal capacity, and reflects oil flow and cooling effects. The initial model parameters of the winding lumped parameter thermal circuit model are determined based on transformer structural parameters, factory test data, type test data, and / or historical operating data, and can be corrected during operation by the model version management unit based on online residuals and parallel verification results.

[0089] In this embodiment, the thermal state prediction unit maintains a thermal state vector X(k), which includes the winding hot spot temperature state variable and the equivalent thermal state variable corresponding to the region where each temperature measuring point is located.

[0090] The thermal state prediction unit employs a lumped-parameter thermal circuit model of the windings based on the thermoelectric analogy principle. This lumped-parameter thermal circuit model equates the heat conduction, convection, and radiation processes of the transformer to a thermal conduction network that includes nonlinear thermal resistance, thermal capacity, and reflects oil flow and cooling effects. The initial model parameters of the lumped-parameter thermal circuit model are based on the transformer structural parameters, factory test data, type test data, and / or the model parameter set read by the thermal state prediction unit from the current stable model version. The model parameter set includes at least the thermal time constant, thermal resistance coefficient, thermal capacity coefficient, and cooling efficiency coefficient. The thermal time constant, thermal resistance coefficient, and thermal capacity coefficient characterize the heat transfer relationship between the windings, oil channels, and adjacent structural components, while the cooling efficiency coefficient characterizes the correction effect of the cooling equipment on the thermal conduction or state transition process after its activation.

[0091] The thermal state prediction unit employs a lumped-parameter thermal circuit model of the winding based on the thermoelectric analogy principle. This lumped-parameter thermal circuit model equates the heat conduction, convection, and radiation processes of the transformer to a thermal conduction network that includes nonlinear thermal resistance, thermal capacity, and reflects oil flow and cooling effects. The initial model parameters of the lumped-parameter thermal circuit model are calculated based on transformer structural parameters, factory test data, type test data, and / or using state transition equations to obtain prior predictions of the thermal state for the current time window.

[0092]

[0093] in, This represents the predicted prior state value for the current time window. This is a posterior estimate of the thermal state for the previous time window. The state transition matrix is ​​determined by the thermal time constant. For the input matrix, The input vector for operational parameters includes at least the square of the equivalent load current, the top oil temperature, and the ambient temperature. When the cooling equipment is activated, the cooling state affects the input matrix through the cooling efficiency coefficient. or state transition matrix .

[0094] Furthermore, a predicted temperature vector for each measuring point is calculated based on the measurement mapping relationship, wherein the measurement mapping matrix maps the equivalent thermal state in the state vector to the predicted temperature value at each measuring point location. The prior value of the hot spot temperature is extracted from the state vector by the extraction vector.

[0095] Step S4: Calculate the residuals and assess the health status and sensitivity of the measurement points.

[0096] Specifically, the measurement point status evaluation unit calculates the residual between the measured temperature and the predicted temperature value of each effective measurement point, and the residual is obtained by subtracting the predicted temperature from the measured temperature.

[0097] Furthermore, the measurement point status evaluation unit maintains a continuous health status factor for each measurement point, with a value between 0 and 1.

[0098] When the residual amplitude of a measuring point continuously exceeds the upper limit of the residual allowance over multiple consecutive time windows, a deduction is applied to the health status factor. The upper limit of the residual allowance can be set as 2 to 3 times the standard deviation of the residual during the measuring point's historical normal operation, or it can be set as a fixed temperature value such as 3°C to 8°C.

[0099] When the rate of change of temperature at the measuring point exceeds the thermal inertia constraint boundary and has no reasonable correlation with load changes, oil temperature changes, and cooling state changes, a deduction is applied to the health state factor.

[0100] When the direction of temperature difference change between the measuring point and adjacent measuring points in the same zone is mismatched with the direction of load change, a deduction is applied to the health status factor.

[0101] The deduction step size can be set to 0.05 to 0.20.

[0102] Once all three abnormal conditions mentioned above have disappeared within multiple consecutive time windows, a recovery operation is applied to the health status factor. The recovery step size can be set to, for example, 0.02 to 0.10, which is lower than the deduction step size.

[0103] Furthermore, the measuring point status evaluation unit calculates the sensitivity of the hotspot temperature prior estimate to the temperature of each measuring point. The sensitivity is defined as the absolute value of the partial derivative of the hotspot temperature prior value with respect to the measuring point temperature value. A higher sensitivity indicates a greater contribution of the measuring point to the hotspot temperature estimation. The sensitivity can also be determined by the correlation between the measuring point temperature and the winding hotspot temperature verification value in historical data.

[0104] Step S5: Decompose the importance of the measurement point into observation importance and verification importance. Observation importance is negatively correlated with the residual, while verification importance is positively correlated with the residual.

[0105] See Figure 3 As shown, the measurement point importance calculation unit receives residuals, health status factors, and sensitivity from the measurement point status evaluation unit as inputs.

[0106] Specifically, the residual amplitude and sensitivity are normalized to obtain normalized residuals and normalized sensitivity.

[0107] The normalized residual is calculated by dividing the residual amplitude by the sum of the normalized residual reference value and the residual amplitude. The normalized residual reference value can be set as 1.5 to 3 times the average residual value during the historical normal operation of the measuring point.

[0108] The normalized sensitivity is calculated by dividing the sensitivity by the sum of the normalized sensitivity reference value and the sensitivity. The normalized sensitivity reference value can be, for example, the historical mean or median of the sensitivity at that measurement point.

[0109] Furthermore, the measurement point importance calculation unit calculates the observation importance using the following formula:

[0110]

[0111] in, For measuring points The importance of observation, For measuring points Health status factors For normalization sensitivity, The residuals are normalized. This calculation method ensures that health-sensitive monitoring points with smaller residuals have higher observation importance, while monitoring points with larger residuals or deteriorating health status have observation importance approaching 0. This observation importance is used to determine the participation weight of the corresponding monitoring point in subsequent hotspot temperature fusion estimation.

[0112] Accordingly, the measurement point importance calculation unit calculates the verification importance, and the calculation formula is as follows:

[0113]

[0114] in, For measuring points The importance of the review The preset compensation coefficient ranges from 0.3 to 0.8. The first term in the formula ensures that health-sensitive but disability-abnormal measurement points receive higher review importance, while the second term ensures that critical measurement points that are sensitive but have deteriorating health status still maintain a certain level of review attention. The review importance is used to determine the sampling review frequency of the corresponding measurement point in the next time window.

[0115] For measurement points that are health-sensitive and have normal residuals, their observation importance is high while their verification importance is low, and they should be included in sampling and estimation at a higher frequency. For measurement points that are health-sensitive but have a sudden increase in residuals, their observation importance is low while their verification importance is high, and their weight should be suppressed in estimation, but they should be sampled and verified at a higher frequency to determine the nature of the anomaly. For measurement points that are in deteriorating health but are location-sensitive, their observation importance approaches zero while their verification importance remains at a moderate level, and they should not be included in estimation, but a certain sampling frequency should be maintained to monitor their recovery. For healthy, stable but low-sensitivity measurement points, both their observation importance and verification importance are low, and the sampling frequency can be reduced to save communication and computing resources.

[0116] Step S6: Adjust the sampling schedule based on the observation importance and the verification importance.

[0117] See also Figure 3 As shown, the sampling scheduling unit receives the observation importance and the verification importance, and calculates the sampling interval 309 for each measuring point in the next time window. The formula for calculating the sampling interval 309 is:

[0118]

[0119] in, For measuring points sampling interval, As the reference sampling interval, and These are positive weighting coefficients. The resulting quotient is limited to the range between the minimum sensor response interval and the maximum monitoring requirement interval.

[0120] The baseline sampling interval can be set to, for example, 30 seconds. The minimum sensor response interval can be set to, for example, 5 to 10 seconds. The maximum monitoring interval can be set to, for example, 120 to 300 seconds. The weighting coefficient... and For example, the values ​​can be set to 0.5 and 1.0 respectively, or adjusted according to the actual sampling resource configuration.

[0121] When the field sensors or data acquisition unit do not support independent adjustment of the sampling period per measurement point, the sampling scheduling unit adjusts the sampling scheduling command to increase the frequency of high-importance measurement points in the polling sequence, effectively shortening their effective sampling interval. Specifically, a master station time-sharing scheduling and priority queue mechanism is adopted. Under RS-485, Modbus, or other fieldbus communication protocols, more polling time slots are allocated to high-importance measurement points to achieve non-equal-interval polling without communication conflicts. When multiple high-priority measurement points compete for the same time slot, they are arranged in descending order of importance, and low-priority measurement points can temporarily relinquish time slots.

[0122] Step S7: The candidate model can only replace the stable model after it has been validated in parallel.

[0123] See Figure 4 As shown, the model version management unit determines whether the model parameter update permission is valid. The model parameter update permission is valid when the data involved in the update is complete and valid, the health status factors of all participating measurement points are higher than the minimum confidence threshold, the number of participating measurement points is not less than a preset lower limit and their spatial distribution covers the main area of ​​the winding, and the minimum singular value of the matrix synthesized from the weight matrix constructed by the measurement point observation importance and health factors and the model output sensitivity matrix to parameters is not lower than the information content threshold.

[0124] The minimum confidence threshold can be set to 0.6 to 0.8. The preset lower limit for the number of measurement points participating in the update can be set to 50% to 60% of the total number of measurement points, and no less than 3.

[0125] When the model parameter update permission is granted, the model version management unit, aiming to reduce the weighted sum of squared residuals, modifies the model parameters of the current stable model version using a sensitivity matrix and regularization terms, generates candidate model parameters, and restricts the updated parameters to a physically feasible range defined by the thermal time constant, thermal resistance, and thermal capacity. If a parameter exceeds the feasible range boundary, it is adjusted to the closest boundary value to ensure that the candidate model parameters conform to the physical characteristics of a transformer.

[0126] Furthermore, the model version management unit enables the candidate model and the stable model to operate in parallel within multiple consecutive time windows. Within each time window, the weighted residual evaluation values ​​of the stable model and the candidate model at each valid measurement point are calculated, and the weighted residual evaluation values ​​are normalized with the importance of the measurement point observations as weights.

[0127] When the observation importance-weighted residual of a candidate model remains lower than that of the stable model for a consecutive preset number of time windows, and the hotspot estimates and alarm levels do not exhibit abnormal fluctuations, the model version management unit switches the candidate model to a new stable model. The consecutive preset number of time windows can be, for example, 5 to 20. Abnormal fluctuations refer to changes in the hotspot estimates exceeding a preset fluctuation range between two consecutive time windows, which can be, for example, 2°C to 5°C. If the candidate model does not meet the above switching conditions, the candidate model is discarded, and the original stable model remains unchanged.

[0128] Throughout the parallel verification process, the degradation estimation and alarm output unit uses only the hotspot temperature estimate output by the stable model version as the basis for temperature rise alarms and external outputs.

[0129] Step S8: Evaluate the cooling effect after the cooling system is put into operation and provide feedback for correction.

[0130] See Figure 5 As shown, when the cooling equipment switches from being deactivated to being activated, the cooling feedback evaluation and correction unit establishes a cooling feedback window that includes thermal inertia delay time and effect evaluation duration. The thermal inertia delay time can be set to, for example, 3 to 10 minutes, and the effect evaluation duration can be set to, for example, 10 to 30 minutes.

[0131] Furthermore, the cooling feedback evaluation and correction unit calculates the temperature rise rate before the cooling action and the temperature rise rate after the cooling action, and the difference between the two is the improvement amount of the temperature rise rate.

[0132] Accordingly, the squared change in load and the change in oil temperature before and after the cooling action are calculated.

[0133] The cooling effect evaluation quantity is obtained by subtracting the load square change correction term and the oil temperature change correction term from the improvement in the temperature rise rate. The load change correction coefficient can be set to, for example, 0.001 to 0.01℃ / (min·A²), and the oil temperature change correction coefficient can be set to, for example, 0.02 min. -1 up to 0.15 min -1 .

[0134] When the cooling effect evaluation value is lower than the effective threshold, the cooling efficiency coefficient is reduced or the cooling recommendation trigger temperature threshold is lowered by a preset step size. The effective threshold can be set to, for example, 0.1℃ / min to 0.5℃ / min. The reduction range of the cooling recommendation trigger temperature threshold can be set to, for example, 1℃ to 3℃. The corrected cooling efficiency coefficient is fed back into the thermal state prediction model in step S3, forming a closed-loop correction.

[0135] If the cooling status of each group can be obtained, a cooling feedback window is established for each cooling group and the cooling effect evaluation quantity is calculated for each group. The cooling efficiency coefficient of each group is then corrected accordingly.

[0136] Step S9: Reduce the observation importance of abnormal measurement points, perform downgrade estimation, and output alarm.

[0137] Specifically, when the health status factor of the monitoring point is lower than the abnormal threshold, the importance of the observation is calculated using the formula in step S5. ,because If the value is significantly reduced, the observation importance of the measuring point will automatically approach zero. The anomaly threshold can be set to, for example, between 0.3 and 0.5.

[0138] When the observation importance approaches zero, the measuring point does not directly participate in the hotspot temperature fusion estimation. The downgraded estimate is determined by a weighted combination of the thermal model prediction, the temperature value of adjacent healthy measuring points weighted by spatial distance, and the oil temperature and historical trend extrapolation value, with the addition of a downgraded monitoring status identifier and a data source identifier.

[0139] The information output by the degradation estimation and alarm output unit includes at least: the estimated value of the winding hot spot temperature, the temperature rise of the winding relative to the oil temperature, the measured temperature and predicted temperature of each measuring point, the health status factor of each measuring point, the observation importance and verification importance of each measuring point, the current sampling scheduling strategy, the current model version status identifier, the cooling effect evaluation result, the cooling intervention suggestion or control command, the degradation estimate value and data source identifier of the abnormal measuring point, and the temperature rise alarm and the abnormal measuring point alarm.

[0140] When the health status factor of an abnormal measurement point returns to the normal range, the residual returns to within the upper limit of the residual allowable range, and the spatial consistency with adjacent measurement points returns to normal within multiple consecutive time windows, its normal participation status is gradually restored.

[0141] In one specific implementation, when communication with the host computer is interrupted, the system continues to perform all monitoring, estimation, and alarm functions locally, and caches all thermal event records, model update records, cooling feedback records, and alarm records in local storage. When communication is restored, the cached data is retransmitted in chronological order to ensure the continuity of monitoring records.

[0142] Example 3 Verification

[0143] A prototype was deployed on an in-operation oil-immersed transformer in a 110kV substation for continuous operation testing. The prototype includes eight PT100 platinum resistance temperature sensors, respectively located at the upper and lower parts of the high-voltage winding, the upper and lower parts of the low-voltage winding, the top oil temperature measurement point, the bottom oil temperature measurement point, the ambient temperature measurement point, and a blind temperature measuring tube on the tank wall. Additionally, two fiber Bragg grating temperature sensors are deployed near hot spots inside the windings. The measured values ​​from these fiber Bragg grating temperature sensors are used as the true reference for the winding hot spot temperature for verification. The core processing unit uses an embedded processing board based on ARM Cortex-A72, and the monitoring results are transmitted via the IEC 61850 protocol.

[0144] Typical operating condition 1: Sudden load increase condition

[0145] During peak summer load periods, the transformer's equivalent load current increased from 62% to 87% of the rated current within approximately 45 seconds. The thermal event time window management unit identified the load change thermal event and established a time window. Prior to the event, the estimated winding hotspot temperature was approximately 78.3℃. The residual at the 5th measuring point expanded from the normal range to approximately 4.2℃, the health status factor decreased from 0.94 to 0.81, the observation importance decreased from 0.78 to 0.31, the review importance increased from 0.12 to 0.56, and the sampling interval shortened from 30 seconds to approximately 12 seconds. Twelve minutes after the event, the estimated hotspot temperature reached approximately 96.5℃, and the system issued a temperature rise warning. Subsequently, the residual at the 5th measuring point decreased, and the health status factor and observation importance gradually recovered.

[0146] Typical operating condition 2: Cooler in operation and effectively reducing temperature

[0147] The transformer operated at approximately 82% load, with the estimated hot spot temperature fluctuating between 92°C and 95°C. After the cooler was activated, a cooling feedback window was established with an 8-minute delay and a 20-minute evaluation period. The improvement in the rate of temperature rise was 0.41°C / min, and the cooling effect evaluation was approximately 0.39°C / min, exceeding the effective threshold of 0.2°C / min. The system determined the cooling to be effective, and the cooling efficiency coefficient was increased from 0.72 to 0.78. Approximately 25 minutes later, the estimated hot spot temperature dropped back to approximately 84°C.

[0148] Typical operating condition 3: Cooler in operation but with insufficient effect

[0149] The transformer was operating at approximately 78% load. After the cooler was put into operation, the improvement in temperature rise rate was only 0.07℃ / min, and the cooling effect evaluation was approximately 0.055℃ / min, both below the effective threshold. The system determined that the cooling effect was insufficient, lowered the cooling efficiency coefficient, and output a check prompt. Subsequent inspection revealed that a loose fan belt was causing insufficient airflow.

[0150] Typical operating condition 4: Automatic degradation estimation condition due to abnormal measuring points

[0151] The residual at measuring point 7 suddenly increased to approximately 9.8℃, with a temperature change rate of approximately 7.2℃ / min, exceeding the theoretical upper limit, and adjacent measuring points did not change synchronously. Since this rate of change could not be attributed to load, oil temperature, or cooling changes, the system directly marked this measuring point as abnormal and triggered measuring point abnormality handling, without establishing a global thermal event time window. After three consecutive time windows, the system reduced the health status factor of measuring point 7 to 0.22, using the observation importance calculation formula. The system automatically reduced the observation importance to below 0.03 and switched to degraded estimation mode. During degraded estimation, the hotspot temperature estimate fluctuated within ±1.8℃. Maintenance personnel found that the sensor wiring terminals were loose; after tightening, the measuring point returned to normal, and the health status factor gradually recovered to 0.91, allowing it to participate in estimation again.

[0152] The above are merely preferred embodiments 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 on-line monitoring of temperature rise in transformer windings based on multi-point temperature sensing, characterized in that, include: Collect multiple temperature values ​​and operating data of the transformer; Establish hot event time windows to align multi-source data; The predicted temperature and the prior hot spot temperature are calculated based on the lumped parameter thermal circuit model of the winding. Calculate the residuals and assess the health status and sensitivity of the measuring points; The importance of measurement points is divided into observation importance and verification importance. The observation importance is negatively correlated with the residuals, and the verification importance is positively correlated with the residuals. The sampling schedule is adjusted based on the observation importance and the verification importance. Candidate models can only replace stable models after being validated in parallel. Evaluate the cooling effect after the cooling system is put into operation and provide feedback for correction. After reducing the observation importance of abnormal measurement points, a downgraded estimate is performed and an alarm is output.

2. The method of claim 1, wherein, The establishment of thermal event time windows to align multi-source data includes: When the load current change exceeds the load threshold, or the cooling equipment status changes, or the temperature rise rate of the measuring point continuously exceeds the temperature rise rate threshold, the earliest time that meets the conditions is taken as the starting point of the thermal event, and the first preset time is extended forward and the second preset time is extended backward to form a thermal event time window. When the rate of change of temperature at a measuring point exceeds the thermal inertia constraint boundary and there is no load, oil temperature or cooling change attributable to it, the measuring point is marked as abnormal and the measuring point abnormality handling is triggered, rather than being used as a trigger condition for establishing a thermal event time window. Within the thermal event time window, the switch data is processed by holding the most recently valid state, and the continuous data is processed by linear interpolation or holding the most recently valid data. Data with severe missing data or time stamp exceeding the limit is marked as invalid.

3. The method of claim 1, wherein, The calculation of predicted temperature values ​​and prior hotspot temperatures based on the lumped-parameter thermal circuit model of the winding includes: A lumped-parameter thermal circuit model for windings, established based on thermoelectric analogy, is used. This model includes nonlinear thermal resistance, thermal capacity, and thermal conductivity reflecting oil flow and cooling effects. The initial model parameters of the lumped-parameter thermal circuit model are determined based on transformer structural parameters, factory test data, type test data, and / or historical operating data. The state transition equation of the lumped-parameter thermal circuit model is recursively derived from the posterior estimate of the thermal state of the previous time window, the square of the current load current, oil temperature, ambient temperature, and the state of the cooling equipment corrected by the cooling efficiency coefficient. The predicted temperature values ​​at each measuring point are calculated from the predicted thermal state values, and the prior values ​​of the winding hot spot temperatures are extracted.

4. The method of claim 1, wherein, The method of decomposing the importance of measurement points into observation importance and verification importance, wherein the observation importance is negatively correlated with the residuals and the verification importance is positively correlated with the residuals, includes: The residual amplitude and sensitivity are normalized respectively to obtain the normalized residual and normalized sensitivity; The observation importance is determined by the product of the health status factor, the normalized sensitivity, and 1 minus the normalized residual, so that the health-sensitive measurement point with the smaller residual has a higher observation importance, which is used to determine the participation weight of the measurement point in the hotspot temperature fusion estimation. The importance of the review is determined by the sum of the first component and the second component. The first component is the product of the health status factor, the normalized sensitivity, and the normalized residual. The second component is the product of the preset compensation coefficient, the difference between 1 and the health status factor, and the normalized sensitivity. This results in a higher review importance for sensitive measurement points with abnormal residuals or key measurement points with deteriorating health, which is used to determine the sampling review frequency of the measurement point.

5. The method of claim 1, wherein, The adjustment of sampling scheduling based on the observation importance and the verification importance includes: The sampling interval for each measuring point in the next time window is determined by the quotient obtained by dividing the baseline sampling interval by 1 and the sum of the linear combination of the observation importance and the verification importance, and the quotient is limited to the minimum response interval of the sensor and the maximum interval of the monitoring requirement. When the sensor does not support independent adjustment of the sampling frequency by measurement point, the sampling schedule is adjusted to increase the frequency of high-importance measurement points in the polling sequence. Through the master station's time-sharing scheduling and priority queue mechanism, conflict-free non-equal interval polling is achieved under the bus communication protocol.

6. The method according to claim 1, characterized in that, The candidate model can only replace the stable model after parallel verification, including: When the data involved in the update is complete and valid, the health status factors of the measurement points involved in the update are all higher than the minimum confidence threshold, the number of measurement points involved in the update is not less than the preset lower limit and the spatial distribution covers the main area of ​​the winding, and the minimum singular value of the matrix synthesized by the weight matrix constructed from the importance of measurement point observation and health factors and the model output sensitivity matrix to parameters is not lower than the information content threshold, the model parameter update is deemed to be valid. With the goal of reducing the weighted sum of squared residuals, candidate model parameters are generated using the sensitivity matrix and regularization term. The candidate model parameters are then restricted to a physically feasible range determined by the thermal time constant and thermal resistance and heat capacity. If the parameters exceed the boundary of the feasible range, they are adjusted to the boundary values. The candidate model and the stable model are run in parallel over multiple consecutive time windows. When the observation importance weighted residual of the candidate model is consistently lower than that of the stable model and the hot spot estimate and alarm level do not fluctuate abnormally, the candidate model is switched to a new stable model. Otherwise, the candidate model is discarded and the original stable model is maintained.

7. The method according to claim 1, characterized in that, The evaluation of cooling effect and feedback for correction after cooling is implemented includes: When the cooling equipment is switched from being deactivated to being activated, a cooling feedback window is established that includes thermal inertia delay time and effect evaluation time. Calculate the improvement in temperature rise rate, load square change, and oil temperature change before and after the cooling action. Subtract the load square change correction term and oil temperature change correction term from the improvement in temperature rise rate to obtain the cooling effect evaluation quantity. When the cooling effect evaluation value is lower than the effective threshold, the cooling efficiency coefficient is lowered by a preset step size or the cooling recommendation trigger temperature threshold is reduced.

8. The method according to claim 1, characterized in that, The process of downgrading the importance of abnormal measurement points and outputting alarms after reducing their observation importance includes: When the health status factor of a measuring point is lower than the abnormal threshold, the observation importance of the measuring point is automatically made close to zero through the calculation method of the observation importance, and it does not directly participate in the hot spot temperature fusion estimation. The degradation estimate is determined by a weighted combination of the thermal model prediction, the temperature value of adjacent healthy monitoring points weighted by spatial distance, the oil temperature and the historical trend extrapolation value, and is accompanied by a degradation monitoring status indicator. When the health status factor of the abnormal measurement point returns to normal, the residual returns to the normal range, and the spatial consistency is restored within multiple consecutive time windows, its normal participation status is gradually restored.

9. A transformer winding temperature rise online monitoring system based on multi-point temperature sensing, characterized in that, It includes a multi-point temperature acquisition unit, an operational data acquisition unit, a thermal event time window management unit, a thermal state prediction unit, a measurement point status evaluation unit, a measurement point importance calculation unit, a sampling scheduling unit, a model version management unit, a cooling feedback evaluation and correction unit, and a degradation estimation and alarm output unit; The output terminals of the multi-point temperature acquisition unit and the operation data acquisition unit are connected to the input terminal of the thermal event time window management unit, and are used to transmit the acquired temperature values ​​and operation data to the thermal event time window management unit for time calculation. The output of the thermal event time window management unit is connected to the input of the thermal state prediction unit, and is used to provide it with time-aligned data. The output terminals of the thermal state prediction unit and the thermal event time window management unit are respectively connected to the input terminal of the measuring point state evaluation unit. The measuring point state evaluation unit is used to calculate residuals, evaluate health status factors, and determine sensitivity. The output of the measuring point status evaluation unit is connected to the input of the measuring point importance calculation unit, which is used to generate observation importance and verification importance. The output of the measurement point importance calculation unit is connected to the input of the sampling scheduling unit and the degradation estimation and alarm output unit, respectively. The sampling scheduling unit is used to adjust the sampling scheduling, and the degradation estimation and alarm output unit is used to correct the hot spot temperature and output an alarm. The model version management unit is bidirectionally connected to the thermal state prediction unit, and is used to load model parameters and perform parallel verification. The cooling feedback evaluation and correction unit is bidirectionally connected to the thermal state prediction unit and is used to evaluate the cooling effect and correct the cooling parameters.

10. The system according to claim 9, characterized in that, The model version management unit is also used to perform parallel operations on candidate models and stable models within multiple consecutive time windows. When the observation importance weighted residual of the candidate model is consistently lower than that of the stable model and the hotspot estimate and alarm level do not fluctuate abnormally, the candidate model is switched to a new stable model. The alarm output is based solely on the hotspot temperature estimate output by the stable model version.

Images