An adaptive update method for gas turbine compressor characteristic curves

By constructing a family of dimensionless reference curves and inverting degradation weights using a digital twin system, and combining this with loss coefficient surfaces to adaptively update the characteristic curves of gas turbine compressors, the problems of non-real-time updates and insufficient accuracy in existing technologies are solved, achieving efficient and reliable characteristic curve updates.

CN123109871APending Publication Date: 2026-11-17XIAN THERMAL POWER RES INST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611217597.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-12
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing gas turbine compressor characteristic curve update technologies cannot keep up with the dynamic degradation state during unit operation in real time, resulting in data lag and insufficient accuracy. In particular, online fitting methods lack support for degradation mechanisms and cannot reflect the differences in the impact of different degradation types under different operating conditions.

Method used

A family of dimensionless pressure ratio and efficiency benchmark curves based on the health status calibration of the gas turbine is constructed. The degradation weight is inverted through the digital twin system, and the point-by-point correction is performed under all operating conditions in combination with the pre-calibrated loss coefficient surface. A credibility verification mechanism is introduced to generate a family of real-time characteristic curves.

Benefits of technology

It achieves mechanism-driven adaptive updating of full-condition characteristic curves without relying on online full 3D simulation, reduces correction errors, ensures the reliability and consistency of update results, and meets the real-time requirements of online monitoring.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123109871A_ABST
    Figure CN123109871A_ABST
Patent Text Reader

Abstract

This invention provides an adaptive update method for gas turbine compressor characteristic curves, belonging to the field of gas turbine digital twin and performance monitoring technology. It can at least partially solve the problems of existing online fitting methods, such as lack of degradation mechanism support, insufficient overall scaling correction accuracy, and poor reliability due to the absence of a quality verification mechanism. The invention includes: constructing a family of dimensionless pressure ratio and efficiency benchmark curves; comparing measured performance with the previous version of the valid curves to determine characteristic anomalies; inputting measurement point data into the digital twin system to invert degradation weights; calling a pre-calibrated loss coefficient surface combined with degradation weights to calculate the full-condition performance correction; generating a real-time characteristic curve family through point-by-point correction and verifying its reliability; and scheduling the update timing according to the unit's operating type. This invention achieves mechanism-driven full-condition point-by-point correction and quality closed-loop verification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of digital twin and performance monitoring technology for gas turbines, specifically to an adaptive update method for gas turbine compressor characteristic curves. Background Technology

[0002] The compressor characteristic curve is a core aerodynamic diagram describing the changes in pressure ratio and efficiency with flow rate and speed. It serves as the foundational data for gas turbine control system parameter tuning, performance degradation assessment, and fault early warning analysis. During long-term service of heavy-duty gas turbines, degradation phenomena such as fouling, tip clearance wear, and blade erosion occur in the flow passage components, leading to an overall shift and morphological distortion of the compressor characteristic curve. If performance calculations are continuously performed using the factory baseline curve, calculation deviations will gradually develop, ultimately causing problems such as control strategy mismatch, distorted energy consumption assessment, and delayed fault early warning.

[0003] Existing compressor characteristic curve update technologies mainly fall into two categories: offline correction and online fitting. The offline correction approach relies on full 3D CFD simulation or bench disassembly tests to calibrate the characteristic curves. This method offers high accuracy, but a single full-condition calculation takes several hours to days, and can only be performed periodically offline. It cannot keep up with the dynamic degradation state during unit operation, resulting in a persistent data lag. The online fitting approach uses real-time measurement data to perform overall translation or scaling fitting on the baseline curve. This method offers better real-time performance but suffers from three core drawbacks: the purely data-driven fitting method does not consider the physical mechanism of degradation, and can only correct single-point deviations near the rated operating condition, failing to reflect the differences in the impact of different degradation types at different flow rates and operating levels; it generally uses a single scaling factor to uniformly correct the entire curve, failing to reflect the differentiated impact of degradation on the curve shape; and it only outputs the correction results without validity verification. When measurement points malfunction or inversion anomalies occur, erroneous results directly contaminate the characteristic curve library, causing cascading errors in subsequent calculations.

[0004] In recent years, digital twin technology for gas turbines has achieved the ability to invert the degradation state of flow path components. However, a complete mapping link from degradation weights to full-condition characteristic curves has not yet been established, and degradation perception capabilities have not been transformed into directly accessible characteristic curve data assets. Therefore, how to establish a mechanism-driven, full-condition point-by-point correction method for compressor characteristic curves, with closed-loop quality verification capabilities, without relying on online full-3D simulation, is a pressing technical problem that needs to be solved in the field of online performance monitoring of gas turbines. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art, and provides an adaptive update method for the characteristic curve of a gas turbine compressor.

[0006] To achieve the above objectives, the present invention provides an adaptive update method for the characteristic curve of a gas turbine compressor, comprising: Step S1: Construct a family of dimensionless pressure ratio and efficiency benchmark curves based on the gas turbine health status calibration; Step S2: Collect measurement point data under steady-state conditions and convert it to a dimensionless coordinate system. After outlier quality control, compare the current measured performance data with the previous version of the valid curve under the same dimensionless conditions. If the deviation exceeds the preset anomaly threshold, mark it as needing to be updated. Step S3: Based on the update requirements, input the quality-controlled measurement point data into the digital twin system to invert and obtain the weight coefficient set corresponding to each of the multiple types of degradation; Step S4: Call the pre-calibrated loss coefficient surfaces corresponding to each type of degradation, and calculate the performance correction amount of all discrete points of the benchmark curve family in combination with the set of weight coefficients. The loss coefficient surfaces characterize the variation law of the relative performance loss rate under unit degradation intensity with the operating parameters. Step S5: Apply the performance correction amount to each discrete point of the reference curve family one by one to generate a real-time characteristic curve family, and calculate the curve reliability based on the current measured operating point. If the reliability is qualified, output the real-time characteristic curve family; otherwise, retain the previous version of the valid curve.

[0007] Further, in step S1, the family of reference curves constructs multiple characteristic lines with the corrected dimensionless flow rate as the horizontal axis and the characteristic line parameters as the vertical axis. Each characteristic line contains multiple discrete flow points. The characteristic line parameters are the corrected rotational speed or the inlet guide vane opening. When the unit is a heavy-duty gas turbine operating at a constant speed in grid connection, the characteristic line parameters are the inlet guide vane opening. The inlet guide vane opening covers multiple settings within the grid-connected operation range, and each opening corresponds to an independent characteristic curve.

[0008] Furthermore, the corrected dimensionless flow rate With corrected speed The calculation formula is: ; ; ; in, To correct the mass flow rate, To measure the mass flow rate, This refers to the total temperature at the compressor inlet. This is the total pressure at the compressor inlet. and These are the ISO standard reference temperature and reference pressure, respectively. For the actual measured speed, To correct the rotational speed, Adjust the flow rate to suit the design operating conditions. For the corrected dimensionless flow rate; The family of reference curves is invoked via a two-dimensional interpolation function: ; ; in, This refers to the reference pressure ratio value of the reference curve family at the corresponding operating point. This represents the baseline efficiency value of the family of baseline curves at the corresponding operating point. and These are the two-dimensional interpolation functions for pressure ratio and efficiency, respectively. The feature line parameters are... Let be the set of discrete points of the family of reference curves.

[0009] Further, in step S2, the determination conditions for steady-state operating conditions are: load fluctuations within a preset time window do not exceed a preset load fluctuation threshold, speed fluctuations do not exceed a preset speed fluctuation threshold, and inlet / outlet pressure fluctuations do not exceed a preset pressure fluctuation threshold; the preset anomaly thresholds include a pressure ratio deviation threshold and an efficiency deviation threshold. When the deviation of the measured pressure ratio relative to the previous version of the effective curve under the same dimensionless operating conditions reaches the pressure ratio deviation threshold, or the deviation of the measured efficiency relative to the previous version of the effective curve reaches the efficiency deviation threshold, the requirement to be updated is marked; the quality control uses statistical criteria to remove outliers, and determines the current data to be invalid when the change in inlet guide vane opening, load change, or ambient temperature change exceeds the corresponding preset invalidity criterion threshold.

[0010] Further, in step S3, the set of weight coefficients is obtained by inversion using a residual minimization algorithm with sparse constraints, and the objective function is: ; in, This is the optimal degenerate weight vector obtained through inversion. For the degenerate weight vector, For the first Measured values ​​at each measuring point The digital twin system is based on the degradation weight vector The calculated output value, For the first Sensitivity weights for each measurement point For the number of measurement points, This represents the total number of degradation types. For the first The weighting coefficient for class degradation, with a value ranging from 0 to 1. Let L1 norm be the degenerate weight vector. is the sparse constraint coefficient.

[0011] Furthermore, in step S4, the first Degraded pressure ratio loss coefficient With efficiency loss coefficient It is a two-dimensional curved surface, calibrated by offline simulation at unit degradation intensity, and is defined as follows: ; ; in, For the first Degraded pressure ratio loss coefficient, For the first The efficiency loss coefficient of the degradation class and These are the reference pressure ratio and reference efficiency values ​​of the reference curve family at the corresponding operating points, respectively. and The first The compression ratio and efficiency values ​​of the degradation class under unit strength, To correct for dimensionless flow rate, For characteristic line parameters, For the first Weighting coefficients for class degradation.

[0012] Furthermore, the performance correction amount adopts a hierarchical calculation strategy: When the total degradation intensity And single-class degradation weight At that time, the performance correction amount is calculated using linear superposition: ; ; When the degradation intensity exceeds the threshold, a nonlinear compensation term is introduced to correct the multi-degradation coupling effect: ; ; in, This is the pressure ratio correction amount. For efficiency correction. and These are the reference pressure ratio and reference efficiency values ​​of the aforementioned family of reference curves, respectively. This represents the total number of degradation types. For the first The weighting coefficient for class degradation and The first The pressure ratio loss coefficient and efficiency loss coefficient are degraded. The sum of all degradation weights. The preset total degradation intensity threshold, The preset single-class degradation weight threshold, This is a nonlinear compensation term for the pressure ratio. For efficiency nonlinear compensation terms, To correct for dimensionless flow rate, These are the characteristic line parameters.

[0013] Furthermore, in step S5, the point-by-point correction formula is as follows: ; ; in, The pressure ratio value at the corresponding discrete point of the real-time characteristic curve family. This represents the efficiency value of the family of real-time characteristic curves at the corresponding discrete points. The reference pressure ratio value of the reference curve family at the corresponding discrete point. This represents the baseline efficiency value of the family of baseline curves at the corresponding discrete points. This is the pressure ratio correction amount for this discrete point. This is the efficiency correction amount at this discrete point. For the first Corrected dimensionless flow rate at discrete flow points For the first The parameter values ​​of the feature lines are calculated simultaneously, along with the performance drift rate. : ; in, For pressure ratio drift rate, The pressure ratio value of the aforementioned family of real-time characteristic curves. The reference pressure ratio value is the reference curve family.

[0014] Furthermore, the reliability of the curve is determined by the relative error of the pressure ratio. and relative error of efficiency The evaluation and calculation formula is as follows: ; ; in, This is the relative error of the pressure ratio. For relative efficiency error, This is the measured pressure ratio at the current actual operating point. The measured efficiency is the actual efficiency at the current operating point. This refers to the calculated pressure ratio value of the real-time characteristic curve family at the corresponding operating point. This represents the calculated efficiency value of the real-time characteristic curve family at the corresponding operating point. This is the corrected dimensionless flow rate corresponding to the current measured operating point. The characteristic line parameters corresponding to the current measured working point; when Not exceeding the preset pressure ratio confidence threshold and If the reliability does not exceed the preset efficiency reliability threshold, the reliability is deemed acceptable; if it exceeds the threshold, the update is deemed invalid, and the previous version's valid curve is retained.

[0015] Furthermore, after confirming the update requirement, the execution timing of steps S3 to S5 is selected according to the unit's operating type. For stable operating units, the update is performed at the operating point that meets the steady-state conditions, and the interval between two adjacent updates is not less than the preset minimum update interval. For units that frequently start and stop, the measurement point data is temporarily stored, and the curve updates of steps S3 to S5 are performed uniformly after the unit stops. The update results are used as the benchmark for the next operating cycle. The update also includes a benchmark iteration step after water washing: after the unit completes offline water washing and operates stably, the real-time characteristic curve family with qualified reliability is marked as a short-term benchmark after water washing. The single-unit correction factor is calculated to offset the inherent offset caused by manufacturing tolerances, assembly deviations, and irreversible degradation. The short-term benchmark replaces the benchmark curve family as the starting point for degradation correction within a preset validity period.

[0016] The beneficial effects of this invention are as follows: This invention uses a pre-calibrated categorized degradation loss coefficient surface as its core mechanism to correct discrete points of the reference curve family under all operating conditions one by one. Unlike the traditional overall fitting method with a single scaling factor, the differential influence of various degradations under different flow rates and operating levels can be accurately matched through the loss coefficient surface, effectively reducing the average correction error under all operating conditions. This solves the problems of insufficient degradation mechanism support and inadequate accuracy in existing online fitting methods.

[0017] This invention uses a dimensionless corrected coordinate system to construct a unified family of reference curves, eliminating the interference of environmental temperature and pressure fluctuations on curve coordinates from the root. Curve drift only reflects the true degradation rather than the false drift caused by changes in operating conditions, and the evaluation results in different seasons and under different environments are consistent and comparable.

[0018] This invention incorporates a reliability verification mechanism, which reversely verifies the accuracy of the real-time curve by measuring the actual working conditions, automatically identifies and intercepts abnormal update results, and forms a quality closed loop of "calculation-verification-output" to avoid curve distortion caused by measurement point failures and inversion anomalies.

[0019] This invention adopts an architecture of "offline pre-calibration plus online fast calculation", which eliminates the need for online full 3D simulation. The time required for a single full-condition curve update is significantly lower than that of the offline simulation method, thus meeting the real-time requirements of online monitoring. Attached Figure Description

[0020] Figure 1 This is an overall flowchart of the adaptive update method for gas turbine compressor characteristic curves according to an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the mechanism of the core correction step in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and beneficial effects of this application clearer, the following detailed description, in conjunction with the accompanying drawings and specific embodiments, further illustrates this application. It should be understood that the specific embodiments described in this specification are merely for explaining this application and are not intended to limit it.

[0022] The overall technical solution of the adaptive update method for the characteristic curve of the gas turbine compressor of the present invention will be described below.

[0023] See Figure 1 The adaptive update method for the characteristic curve of the gas turbine compressor of the present invention includes steps S1 to S5, and each step will be described in detail below.

[0024] Step S1 involves constructing a family of dimensionless pressure ratio and efficiency reference curves based on the gas turbine's health condition calibration. This family of reference curves serves as the underlying reference for all subsequent correction calculations, and its accuracy directly determines the precision of the final output. This invention employs a modified dimensionless coordinate system to characterize the family of reference curves, uniformly converting flow rate and speed to dimensionless values ​​under ISO standard operating conditions, thereby eliminating the influence of environmental temperature and pressure fluctuations on coordinate positioning during the curve construction stage.

[0025] The baseline curve family is constructed as follows: using the corrected dimensionless flow rate as the horizontal axis and the characteristic line parameter as the vertical axis, a family of pressure ratio curves and a family of efficiency curves containing multiple characteristic lines are constructed, with each characteristic line covering multiple discrete flow points. The characteristic line parameter refers to the operating parameter dimension used to distinguish different characteristic lines. Depending on the unit's operating mode, either the corrected speed or the inlet guide vane opening (IGV) can be selected. The selection of the characteristic line parameter depends on the unit's operating characteristics: for variable speed units or general scenarios requiring coverage of start-up and shutdown processes, the corrected speed is used as the characteristic line parameter, as this method can adapt to full-condition analysis with varying speeds; for constant-speed heavy-duty gas turbines with constant speed after grid connection and load adjustment relying on inlet guide vanes, the inlet guide vane opening is used as the characteristic line parameter, as this method closely matches the actual load adjustment pattern of constant-speed units. The selection of the above two characteristic line parameters is not mutually exclusive and can be configured individually or in combination according to engineering needs.

[0026] The calibration data sources for the family of reference curves include, but are not limited to: new machine acceptance test data, post-overhaul health operation data, and performance test data provided by the manufacturer. All reference data are stored in the form of discrete point sets and are equipped with two-dimensional interpolation functions, which can quickly solve for the corresponding reference pressure ratio and reference efficiency given arbitrary corrected dimensionless flow rate and characteristic line parameter values.

[0027] Step S2 involves characteristic anomaly determination and data preprocessing. The prerequisite for updating the compressor characteristic curve is a substantial shift in aerodynamic characteristics, not merely because the unit has entered steady-state operation. The core logic of this step is "steady-state sampling, anomaly comparison, and demand labeling."

[0028] First, the stability of the operating conditions is continuously monitored during unit operation. When the operating conditions meet the steady-state criteria, corresponding measurement data are collected, including parameters such as compressor inlet and outlet temperatures, inlet and outlet pressures, mass flow rate, inlet guide vane opening, and rotational speed. The collected measurement data are converted to a corrected dimensionless coordinate system according to the ISO standard reference temperature and pressure to obtain the corrected dimensionless flow rate, measured pressure ratio, and measured efficiency corresponding to the current steady-state operating conditions.

[0029] Secondly, outlier quality control is performed on the converted measurement data, and statistical criteria are used to remove abrupt outliers to ensure the reliability of subsequent comparison data. At the same time, it is determined whether the rate of change of the current operating condition exceeds the validity criterion. When the change in the inlet guide vane opening, load change, or ambient temperature change exceeds the corresponding preset invalidity criterion threshold, the current data is determined not to meet the steady-state validity requirements and is not included in the anomaly judgment.

[0030] Then, the measured performance parameters after quality control are compared with the previous version of the valid curve at the same dimensionless operating point, and the relative deviations of pressure ratio and efficiency are calculated respectively. When the deviation exceeds the preset anomaly threshold, it is determined that the compressor characteristics have undergone substantial changes, and an update requirement is generated; when the deviation does not reach the preset anomaly threshold, it is determined that the characteristics have not changed significantly, the previous version of the valid curve is retained, and subsequent update calculations are not initiated.

[0031] Step S3 is the degradation weight inversion. After confirming the update requirements, the quality-controlled steady-state measurement data is input into the digital twin system, and the inversion algorithm is used to solve for the set of weight coefficients corresponding to each type of degradation.

[0032] The digital twin system described in this invention refers to an online simulation system capable of inverting the degradation state of flow passage components. The core task of degradation inversion is to identify and quantify the severity of various degradation types in compressor flow passage components from a limited set of externally measurable parameters. The degradation types output by the inversion include, but are not limited to, fouling, increased tip clearance, and blade defects. Each type of degradation corresponds to a weighting coefficient, ranging from 0 to 1, where 0 represents no degradation and 1 represents that the degradation of that type has reached its limit.

[0033] The inversion algorithm employs a residual minimization method, using the residual between the calculated and measured values ​​of the digital twin model as the optimization objective to search for the degradation weight combination that minimizes the residual. To ensure that the inversion results conform to the physical law in engineering practice that "a few degradation types dominate," a sparsity constraint is introduced into the algorithm. This involves applying an L1 norm penalty term to the degradation weight vector, guiding the inversion results towards a direction where "dominant degradation types are prominent and secondary degradation types are weakened." The physical significance of the sparsity constraint lies in the fact that in actual gas turbine operation, various degradation types typically do not develop uniformly, and inversion without constraints can easily produce non-physical solutions where multiple degradation types are evenly distributed.

[0034] The aforementioned residual minimization methods can be implemented in, but are not limited to, the following specific ways: least squares method with L1 norm regularization, elastic network regularization method, sequential quadratic programming solution method, or hybrid solution method of genetic algorithm and gradient optimization. All of these implementations aim to minimize the residual at the measurement points, can introduce sparse constraint terms to ensure the physical rationality of the inversion solution, and output a set of weight coefficients corresponding to multiple types of degradation.

[0035] Step S4 is the point-by-point performance correction under all operating conditions. The core of this step is to call the pre-calibrated loss coefficient surfaces corresponding to various types of degradation, and combine them with the weight coefficient set output in step S3 to calculate the performance correction amount at all discrete points of the benchmark curve family.

[0036] The loss coefficient surface is the core data carrier for the mechanism-driven point-by-point correction of this invention. Each type of degradation corresponds to a pair of loss coefficient surfaces, namely the pressure ratio loss coefficient surface and the efficiency loss coefficient surface, defined as the relative pressure ratio loss rate and relative efficiency loss rate caused by this type of degradation under unit degradation intensity. The loss coefficient surface uses the corrected dimensionless flow rate and characteristic line parameters as two-dimensional independent variables, and its value varies with the location of the operating point, accurately depicting the differentiated influence of the same degradation type under different operating conditions.

[0037] The loss coefficient surface is obtained through pre-calibration via offline simulation. Offline simulation refers to performing full-condition performance simulations under unit strength conditions for various degradation types using computational fluid dynamics (CFD) or equivalent aerodynamic analysis methods to obtain performance difference data before and after degradation, thereby constructing a mapping relationship between the loss coefficient and operating parameters. The loss coefficient surface can also be obtained through calibration using actual degradation test data. Both calibration methods acquire full-condition performance difference data under controlled degradation conditions and establish a two-dimensional mapping relationship between the loss coefficient and operating parameters accordingly.

[0038] The performance correction is calculated using a tiered strategy. When the degradation level is in the mild to moderate range, the coupling effect between different types of degradation is weak, and linear superposition can meet the accuracy requirements with the highest computational efficiency. When the degradation intensity exceeds a preset threshold, the coupling effect between multiple types of degradation becomes significant. A nonlinear compensation term is introduced to correct the additional impact of the coupling effect on performance, ensuring computational accuracy in strong degradation scenarios.

[0039] Step S5 involves real-time curve generation and reliability verification. For each discrete point in the baseline curve family, the pressure ratio correction and efficiency correction calculated in step S4 are applied. These are subtracted point-by-point to generate a real-time characteristic curve family with the same topological structure as the baseline curve family. This point-by-point correction method fully preserves the discrete mesh and morphological features of the baseline curves, producing only performance drift corresponding to degradation, thus avoiding the morphological distortion that might occur with the overall scaling correction method.

[0040] The synchronous output performance drift rate quantifies the performance decline of the real-time characteristic curve relative to the healthy baseline curve, intuitively reflecting the overall degradation degree of the compressor.

[0041] Credibility verification is a crucial step in achieving a quality closed loop in this invention. The verification method involves comparing the measured pressure ratio and efficiency at the current measured operating point with the calculated values ​​of the real-time characteristic curve at the corresponding operating point, calculating the relative error of the pressure ratio and the relative error of the efficiency. When both relative errors do not exceed the corresponding preset credibility threshold, the credibility of this curve update is deemed acceptable, and the real-time characteristic curve family is output as the new valid curve. When either relative error exceeds the preset credibility threshold, the update is deemed abnormal, the previous valid curve is automatically retained, the curve library is not updated, and an error message is output. This credibility verification mechanism effectively intercepts erroneous update results caused by measurement point failures, inversion anomalies, or insufficient data quality, preventing distorted curves from contaminating subsequent calculations.

[0042] Among them, the type-based adaptive update scheduling is a process in which, after confirming the update requirement, the system selects the execution time of steps S3 to S5 according to the unit's operating type, taking into account both update timeliness and operational safety.

[0043] For units operating stably, i.e. those that have been continuously connected to the grid for a long time and have small load fluctuations, after confirming the update requirements, select a typical operating point that meets the steady-state conditions to perform the update, avoiding calculations during load fluctuations or operation adjustments; at the same time, set a preset minimum update interval to avoid repeated updates in a short period of time.

[0044] For units that frequently start and stop, i.e., units with many start-stop cycles and short operating cycles, after confirming the update requirement, the current measurement point data and weight coefficient set and other intermediate results are temporarily stored, and the full curve update is not performed during the operation phase; after the unit is shut down normally, the full curve update calculation of steps S3 to S5 is performed uniformly during the shutdown window, and the update result is used as the calculation benchmark for the next operating cycle.

[0045] The above is a description of the overall technical solution of the present invention. The following detailed explanation of each step is provided through specific embodiments.

[0046] Example 1 This embodiment uses a 9FA heavy-duty gas turbine from a power plant as the application example. This turbine is equipped with a 17-stage axial-flow compressor, and its ISO design parameters are: pressure ratio 15.4, design mass flow rate 645 kg / s, isentropic efficiency 88.5%, and rated speed 3000 rpm. During grid-connected operation, the unit operates at a constant speed, and load regulation relies on the coordinated control of IGV opening and fuel quantity. The commonly used IGV operating range is 42 to 88 degrees Celsius. This embodiment targets a steady-state operation scenario at 85% base load, assuming the unit is operating stably with base load. The IGV opening is used as a characteristic parameter, and a complete update process is executed.

[0047] Step S1: Construct a family of dimensionless pressure ratio and efficiency benchmark curves based on the gas turbine health status calibration.

[0048] The family of reference curves in this embodiment is used to correct dimensionless flow rates. The x-axis represents the IGV opening degree as the characteristic curve parameter. The reference curve family includes 5 IGV opening degree characteristic curves, corresponding to 42 degrees, 55 degrees, 68 degrees, 78 degrees, and 88 degrees, respectively, covering 5 typical opening degrees and the commonly used operating range after grid connection. Each characteristic curve has 20 dimensionless flow discrete points evenly distributed, for a total of 100 discrete points. The reference temperature and pressure are set to... , Design to correct flow .

[0049] Corrected dimensionless flow The calculation process is as follows: First, based on the measured mass flow rate... Total temperature at the compressor inlet and total imported pressure Calculate the corrected mass flow rate using the following formula. : ; Then, divide the corrected mass flow rate by the design condition corrected flow rate to obtain the corrected dimensionless flow rate: ; For variable speed scenarios, correct the speed. The calculation formula is: ; in, This is the actual measured rotational speed.

[0050] All discrete points of the family of benchmark curves in the data set Stored in the form of [format], with a corresponding two-dimensional interpolation function: ; ; in, For the characteristic line parameter, the IGV opening is used in this embodiment. This interpolation function can quickly solve for the corresponding reference pressure ratio given any modified dimensionless flow rate and IGV opening. and benchmark efficiency .

[0051] Step S2: Collect measurement point data under steady-state conditions and convert it to a dimensionless coordinate system. Compare the current measured performance with the previous version of the effective curve under the same dimensionless conditions. If the deviation exceeds the preset anomaly threshold, mark it as needing to be updated.

[0052] Steady-state determination: In this embodiment, the steady-state determination conditions are set as follows: load fluctuation not exceeding ±2%, speed fluctuation not exceeding ±0.5%, and inlet / outlet pressure fluctuation not exceeding ±1% within 5 consecutive minutes. During the continuous and stable operation of the unit at 85% load, the load fluctuation is 1.1%, the speed fluctuation is 0.28%, and the inlet / outlet pressure fluctuation is 0.65% within 5 minutes, all three indicators meet the steady-state determination conditions.

[0053] Data Acquisition and Quality Control: Data from 10 consecutive sampling periods was collected. After removing abrupt outliers using a three-standard-deviation criterion, the mean value was calculated. Simultaneously, the changes in inlet guide vane opening, load, and ambient temperature were checked to ensure they did not exceed 5%. Data was deemed invalid if any of these conditions were exceeded. In this embodiment, none of the three changes exceeded the thresholds, and the data was considered valid.

[0054] The mean parameters after quality control are: inlet total temperature Imported total pressure Total outlet pressure: 1485 kPa; Total outlet temperature: 652 K; Actual flow rate: IGV opening 78 degrees, speed 3005 rpm. The corrected dimensionless flow rate is calculated. The measured pressure ratio is 14.73 and the measured efficiency is 86.1%.

[0055] Anomaly Detection: Compared with the previous version of the effective curve, under the same dimensionless operating conditions, the pressure ratio deviation is 0.92%, exceeding the pressure ratio deviation threshold of 0.5%; the efficiency deviation is 1.83%, exceeding the efficiency deviation threshold of 1.0%. Both deviations exceed the preset anomaly thresholds, indicating that the compressor characteristics have undergone substantial changes, and a requirement for updating is generated.

[0056] Update execution timing determination: This unit is a stable operating unit with basic load. The current 85% load steady-state condition meets the execution conditions. The current operating point is selected to execute this curve update.

[0057] Step S3: Input the quality-controlled measurement point data into the digital twin system to invert and obtain the weight coefficient set corresponding to each of the multiple types of degradation.

[0058] The digital twin system in this embodiment employs a residual minimization algorithm with sparse constraints to perform degenerate weight inversion. The objective function is: ; in, For the first Measured values ​​at each measuring point For digital twin models based on degenerate weight vectors The calculated output value, For the sensitivity weight of the measurement point, For the number of measurement points, This represents the total number of degradation types. For the first The weighting coefficient for class degradation, with a value ranging from 0 to 1. For the sparsity constraint coefficients, in this embodiment .

[0059] The inversion yields three types of degradation weights: fouling Increased blade tip clearance Leaf defects Total degradation intensity .

[0060] Step S4: Call the pre-calibrated loss coefficient surfaces corresponding to each type of degradation, and calculate the performance correction amount of all discrete points of the benchmark curve family in combination with the set of weight coefficients.

[0061] This embodiment uses IGV aperture as the dimension of the feature line parameter. Degenerate pressure ratio loss coefficient surface and efficiency loss coefficient surface Defined as: ; ; in, For IGV opening, and The first The compression ratio and efficiency value of class degradation under unit strength.

[0062] In this embodiment, the total degradation intensity Less than the total degradation intensity threshold Furthermore, the degradation weights of each individual class are all less than the threshold for individual class degradation weights. This falls within the applicable range of linear superposition. The performance correction is calculated using the linear superposition method: ; ; The correction amount is calculated point-by-point for 100 discrete points. Correction result at rated operating point: pressure ratio correction amount. Efficiency correction amount percentage points.

[0063] When the degradation intensity exceeds the above threshold, a nonlinear compensation term is introduced. and To correct the multi-degradation coupling effect, the nonlinear compensation term is calibrated in a manner similar to the loss coefficient surface, and is obtained by the difference between the result of offline simulation under the combined effect of multiple types of degradation and the result of linear superposition.

[0064] Step S5: Apply the performance correction amount to each discrete point of the reference curve family one by one to generate a real-time characteristic curve family, and calculate the curve reliability based on the current measured operating point.

[0065] The point-by-point correction formula is: ; ; in, Number the discrete flow points. The feature lines are numbered. The real-time characteristic curve family generated after point-by-point correction maintains the same discrete grid structure as the baseline curve family.

[0066] Real-time pressure ratio at rated operating point Real-time efficiency .

[0067] Performance drift rate: ; ; Reliability verification: Calculate the relative error of pressure ratio and the relative error of efficiency. ; ; In this embodiment, the preset pressure ratio confidence threshold is 1%, and the preset efficiency confidence threshold is 1.5%. A pressure ratio relative error of 0.6% is less than 1%, and an efficiency relative error of 0.15% is less than 1.5%, both meeting the qualification criteria. Therefore, this update is deemed valid, and the real-time characteristic curve family is output as the new version's valid curves.

[0068] The execution timing of steps S3 to S5 is scheduled according to the unit's operating type.

[0069] In this embodiment, the unit is a stable operating unit with a base load. After confirming the update requirement, the curve update is performed at the 85% load condition point that currently meets the steady-state conditions. The preset minimum update interval is 24 hours, meaning that if less than 24 hours have passed since the last effective update, the update will be temporarily suspended even if a characteristic change is detected.

[0070] The entire curve update took only 1.2 seconds, which is more than 10,000 times faster than the 4 to 6 hours required for offline full 3D CFD simulation, thus meeting the real-time requirements of online monitoring.

[0071] Example 2 This embodiment uses a 9FA gas turbine as an example. This unit operates in a frequent start-stop peak-shaving mode, starting and stopping twice a day with a short operating cycle. This embodiment aims to verify the applicability of the present invention in a frequent start-stop operation mode.

[0072] Step S1: Construct a family of dimensionless pressure ratio and efficiency benchmark curves based on the gas turbine health status calibration.

[0073] The reference curve family is constructed in the same way as in Example 1, using 5 IGV opening characteristic lines at rated corrected speed, with each opening corresponding to 20 dimensionless flow discrete points, for a total of 100 discrete points. The reference temperature and pressure and the design corrected flow parameters are consistent with those in Example 1.

[0074] Step S2: Collect measurement data under steady-state conditions and compare them to determine whether there is a need for updates.

[0075] During operation, the unit load dropped from 70% to 52%, and the operating condition fluctuation exceeded the load fluctuation threshold in the steady-state determination conditions. The system did not collect data or perform anomaly detection. After the unit returned to 55% load and operated stably for 12 minutes, the load fluctuation, speed fluctuation, and pressure fluctuation within the 5-minute window all met the steady-state determination conditions.

[0076] Steady-state parameters collected: Inlet total temperature Imported total pressure Actual flow rate: 458 kg / s; IGV opening: 55 degrees; rotational speed: 2980 rpm. The corrected dimensionless flow rate is calculated from these values. .

[0077] Anomaly detection: Compared with the previous version of the effective curve, the efficiency deviation under the same operating conditions is 1.37%, which exceeds the efficiency deviation threshold of 1.0%, and is marked as requiring an update.

[0078] Step S6: Schedule the execution timing of steps S3 to S5 according to the unit operation type.

[0079] This unit is subject to frequent start-stop operations. The system temporarily stores the current measurement data and does not perform full curve updates online during operation. Once the unit is normally shut down and enters the turning gear state, the system uniformly utilizes computing power to complete the entire process: step S3 (degradation weight inversion), step S4 (performance correction calculation), and step S5 (real-time curve generation and reliability verification). The update results serve as the calculation benchmark for the next day's startup.

[0080] The design intent of this scheduling mode is that during the frequent start-stop phase of a generator unit, operating conditions fluctuate frequently. Performing full-curve updates online not only consumes computing resources but may also be affected by these fluctuations in operating conditions, impacting update accuracy. Utilizing downtime windows to uniformly execute updates ensures update accuracy while avoiding interference with the control logic during operation.

[0081] Anomaly scenario verification: A fault was simulated at a pressure measuring point at an outlet, with the measured value deviating from the true value by 5%. The system calculates the pressure ratio reliability error in step S5. If the pressure ratio exceeds the preset confidence threshold by 1%, the update is automatically deemed invalid, the previous valid curve is retained, and an error message is output. This result verifies that the confidence verification mechanism can effectively intercept erroneous updates caused by measurement point failures.

[0082] Example 3 This embodiment uses the change in characteristic curves of a 9FA gas turbine before and after offline water washing as a scenario to verify the implementation process of the benchmark iteration mechanism after water washing.

[0083] Before water washing: Before water washing, the scale degradation weight of the unit was 0.68 and the efficiency drift rate was -3.2%.

[0084] Step S1: The baseline curve family follows the same construction method as described above.

[0085] Step S2: After water washing, the unit operates stably at 80% load for 15 minutes, meeting the steady-state judgment conditions. Comparing with the previous version of the effective curve, the pressure ratio deviation and efficiency deviation under the same operating conditions both exceed the preset anomaly threshold, indicating a substantial change in the judgment characteristics, and marking it as requiring an update.

[0086] Step S3: Input the degradation weights from the digital twin system. After washing, the scale weight decreased to 0.15, and the scale removal rate was 77.9%. The weights for other degradation types did not change significantly.

[0087] Step S4: Call the IGV dimension loss coefficient surface and combine it with the set of degradation weight coefficients after water washing to calculate the full-condition performance correction.

[0088] Step S5: Point-by-point correction to generate a family of real-time characteristic curves after water washing. Reliability verification results: Pressure ratio relative error 0.52%, efficiency relative error 0.81%, both meeting the qualification requirements.

[0089] Post-wash baseline iteration: The system automatically marks the family of real-time characteristic curves that meet the reliability requirements after washing as the "post-wash short-term baseline". The purpose of the short-term baseline is that: the factory baseline curve family is calibrated based on the manufacturer's standard conditions, while the actual performance of the unit under healthy conditions has an inherent deviation from the factory baseline due to factors such as manufacturing tolerances, assembly gaps, and irreversible corrosion. After washing, the unit is at its optimal performance point close to its healthy state, and the real-time characteristic curves at this time can more accurately reflect the actual performance baseline of the unit.

[0090] The formula for calculating the single-machine correction factor is: ; ; in, and These represent the pressure ratio and efficiency values ​​of a family of real-time characteristic curves that meet reliability standards after water washing. and This refers to the pressure ratio and efficiency values ​​of the factory baseline curve family. This correction factor can offset the inherent deviation between the factory baseline and the actual condition of the machine. Subsequent degradation corrections are based on a short-term baseline after water washing, further improving the accuracy of daily degradation assessments.

[0091] The validity period for short-term benchmarks is set to either 60 days or 800 operating hours, whichever comes first. Upon expiration, the system automatically triggers re-verification and iterative updates. All historical curve versions are stored synchronously, including corresponding operating condition parameters, a set of degradation weight coefficients, and confidence tags, supporting full lifecycle traceability and model iterative optimization.

[0092] In this embodiment, the unit operates in a stable mode, and the first update after water washing is performed directly under 80% load conditions that meet steady-state requirements.

[0093] This invention also includes a supporting system architecture to facilitate the engineering implementation of the aforementioned methods. The system is divided into a data layer, a computation layer, an output layer, and a scheduling layer based on data flow. The data layer is responsible for storing the baseline curve family and acquiring real-time measurement data. The computation layer integrates core computational capabilities for degradation inversion, performance correction, and curve generation. The output layer handles visualization and version management; the visualization output module overlays and displays the baseline curve and real-time curve, annotating drift, characteristic deviation, and reliability indicators. The scheduling layer coordinates the timing of updates based on the unit's operating type. Each module adopts a loosely coupled design, supporting switching between two characteristic line dimensions: corrected rotational speed and IGV opening. It can be directly embedded into existing gas turbine digital twin platforms, reusing their degradation inversion computational power.

[0094] In summary, the embodiments disclosed herein have at least the following technical effects: Using a pre-calibrated categorized degradation loss coefficient surface as the core mechanism, the discrete points of the benchmark curve family under all operating conditions are corrected one by one. Unlike the traditional overall fitting method with a single scaling factor, the differentiated influence of various degradations under different flow rates and different operating levels can be accurately matched through the loss coefficient surface, effectively reducing the average correction error under all operating conditions. This solves the problems of insufficient degradation mechanism support and inadequate accuracy in existing online fitting methods.

[0095] A unified family of reference curves is constructed using a dimensionless corrected coordinate system, which eliminates the interference of environmental temperature and pressure fluctuations on the curve coordinates from the root. The curve drift only reflects the true degradation rather than the false drift caused by changes in operating conditions. The evaluation results in different seasons and under different environments are consistent and comparable.

[0096] The built-in reliability verification mechanism verifies the accuracy of the real-time curve by reverse verification of the measured working points, automatically identifies and intercepts abnormal update results, and forms a quality closed loop of "calculation-verification-output" to avoid curve distortion caused by measurement point failures and inversion anomalies.

[0097] The architecture of "offline pre-calibration plus online fast calculation" eliminates the need for online full 3D simulation. The time required for a single full-condition curve update is significantly lower than that of offline simulation, thus meeting the real-time requirements of online monitoring.

[0098] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for adaptively updating the characteristic curve of a gas turbine compressor, characterized in that, include: Step S1: Construct a family of dimensionless pressure ratio and efficiency benchmark curves based on the gas turbine health status calibration; Step S2: Collect measurement point data under steady-state conditions and convert it to a dimensionless coordinate system. After outlier quality control, compare the current measured performance data with the previous version of the valid curve under the same dimensionless conditions. If the deviation exceeds the preset anomaly threshold, mark it as needing to be updated. Step S3: Based on the update requirements, input the quality-controlled measurement point data into the digital twin system to invert and obtain the weight coefficient set corresponding to each of the multiple types of degradation; Step S4: Call the pre-calibrated loss coefficient surfaces corresponding to each type of degradation, and calculate the performance correction amount of all discrete points of the benchmark curve family in combination with the set of weight coefficients. The loss coefficient surfaces characterize the variation law of the relative performance loss rate under unit degradation intensity with the operating parameters. Step S5: Apply the performance correction amount to each discrete point of the reference curve family one by one to generate a real-time characteristic curve family, and calculate the curve reliability based on the current measured operating point. If the reliability is qualified, output the real-time characteristic curve family; otherwise, retain the previous version of the valid curve.

2. The adaptive update method for gas turbine compressor characteristic curves according to claim 1, characterized in that, In step S1, the reference curve family constructs multiple characteristic lines with the corrected dimensionless flow rate as the horizontal axis and the characteristic line parameters as the vertical markers. Each characteristic line contains multiple discrete flow points. The characteristic line parameters are the corrected rotational speed or the inlet guide vane opening. When the unit is a heavy-duty gas turbine operating at a constant speed in grid connection, the characteristic line parameters are the inlet guide vane opening. The inlet guide vane opening covers multiple settings within the grid-connected operation range, and each opening corresponds to an independent characteristic curve.

3. The adaptive update method for gas turbine compressor characteristic curves according to claim 2, characterized in that, The corrected dimensionless flow With corrected speed The calculation formula is: ; ; ; in, To correct the mass flow rate, To measure the mass flow rate, This refers to the total temperature at the compressor inlet. This is the total pressure at the compressor inlet. and These are the ISO standard reference temperature and reference pressure, respectively. For the actual measured speed, To correct the rotational speed, Adjust the flow rate to suit the design operating conditions. For the corrected dimensionless flow rate; The family of reference curves is invoked via a two-dimensional interpolation function: ; ; in, This refers to the reference pressure ratio value of the reference curve family at the corresponding operating point. This represents the baseline efficiency value of the family of baseline curves at the corresponding operating point. and These are the two-dimensional interpolation functions for pressure ratio and efficiency, respectively. The feature line parameters are... Let be the set of discrete points of the family of reference curves.

4. The adaptive update method for gas turbine compressor characteristic curves according to claim 1, characterized in that, In step S2, the steady-state condition determination criteria are: load fluctuation within a preset time window does not exceed a preset load fluctuation threshold, speed fluctuation does not exceed a preset speed fluctuation threshold, and inlet / outlet pressure fluctuation does not exceed a preset pressure fluctuation threshold. The preset anomaly thresholds include a pressure ratio deviation threshold and an efficiency deviation threshold. When the deviation of the measured pressure ratio relative to the previous effective curve under the same dimensionless operating condition reaches the pressure ratio deviation threshold, or the deviation of the measured efficiency relative to the previous effective curve reaches the efficiency deviation threshold, the requirement to be updated is marked. The quality control uses statistical criteria to remove outliers, and determines the current data to be invalid when the change in inlet guide vane opening, load change, or ambient temperature change exceeds the corresponding preset invalidity criterion threshold.

5. The adaptive update method for gas turbine compressor characteristic curves according to claim 1, characterized in that, In step S3, the set of weight coefficients is obtained by inversion using a residual minimization algorithm with sparse constraints, and the objective function is: ; in, This is the optimal degenerate weight vector obtained through inversion. For the degenerate weight vector, For the first Measured values ​​at each measuring point The digital twin system is based on the degradation weight vector The calculated output value, For the first Sensitivity weights for each measurement point For the number of measurement points, This represents the total number of degradation types. For the first The weighting coefficient for class degradation, with a value ranging from 0 to 1. Let L1 norm be the degenerate weight vector. is the sparse constraint coefficient.

6. The adaptive update method for gas turbine compressor characteristic curves according to claim 1, characterized in that, In step S4, the first Degraded pressure ratio loss coefficient With efficiency loss coefficient It is a two-dimensional curved surface, calibrated by offline simulation at unit degradation intensity, and is defined as follows: ; ; in, For the first Degraded pressure ratio loss coefficient, For the first The efficiency loss coefficient of the degradation class and These are the reference pressure ratio and reference efficiency values ​​of the reference curve family at the corresponding operating points, respectively. and The first The compression ratio and efficiency values ​​of the degradation class under unit strength, To correct for dimensionless flow rate, For characteristic line parameters, For the first Weighting coefficients for class degradation.

7. The adaptive update method for gas turbine compressor characteristic curves according to claim 6, characterized in that, The performance correction amount is calculated using a hierarchical strategy: When the total degradation intensity And single-class degradation weight At that time, the performance correction amount is calculated using linear superposition: ; ; When the degradation intensity exceeds the threshold, a nonlinear compensation term is introduced to correct the multi-degradation coupling effect: ; ; in, This is the pressure ratio correction amount. For efficiency correction. and These are the reference pressure ratio and reference efficiency values ​​of the aforementioned family of reference curves, respectively. This represents the total number of degradation types. For the first The weighting coefficient for class degradation and The first The pressure ratio loss coefficient and efficiency loss coefficient are degraded. The sum of all degradation weights. The preset total degradation intensity threshold, The preset single-class degradation weight threshold, This is a nonlinear compensation term for the pressure ratio. For efficiency nonlinear compensation terms, To correct for dimensionless flow rate, These are the characteristic line parameters.

8. The adaptive update method for gas turbine compressor characteristic curves according to claim 1, characterized in that, In step S5, the point-by-point correction formula is as follows: ; ; in, The pressure ratio value at the corresponding discrete point of the real-time characteristic curve family. This represents the efficiency value of the family of real-time characteristic curves at the corresponding discrete points. The reference pressure ratio value of the reference curve family at the corresponding discrete point. This represents the baseline efficiency value of the family of baseline curves at the corresponding discrete points. This is the pressure ratio correction amount for this discrete point. This is the efficiency correction amount at this discrete point. For the first Corrected dimensionless flow rate at discrete flow points For the first The parameter values ​​of the feature lines are calculated simultaneously, along with the performance drift rate. : ; in, For pressure ratio drift rate, The pressure ratio value of the aforementioned family of real-time characteristic curves. The reference pressure ratio value is the reference curve family.

9. The adaptive update method for gas turbine compressor characteristic curves according to claim 1, characterized in that, The reliability of the curve is determined by the relative error of the pressure ratio. and relative error of efficiency The evaluation and calculation formula is as follows: ; ; in, This is the relative error of the pressure ratio. For relative efficiency error, This represents the measured pressure ratio at the current actual operating point. The measured efficiency is the actual efficiency at the current operating point. This refers to the calculated pressure ratio value of the real-time characteristic curve family at the corresponding operating point. This represents the calculated efficiency value of the real-time characteristic curve family at the corresponding operating point. This is the corrected dimensionless flow rate corresponding to the current measured operating point. The characteristic line parameters corresponding to the current measured working point; when Not exceeding the preset pressure ratio confidence threshold and If the reliability does not exceed the preset efficiency reliability threshold, the reliability is deemed acceptable; if it exceeds the threshold, the update is deemed invalid, and the previous version's valid curve is retained.

10. The adaptive update method for gas turbine compressor characteristic curves according to any one of claims 1 to 9, characterized in that, It also includes a type-based adaptive update scheduling step: after confirming the update requirement, the execution timing of steps S3 to S5 is selected according to the unit's operating type; for stable operating units, the operating point that meets the steady-state conditions is selected for update, and the interval between two adjacent updates is not less than the preset minimum update interval; for units that frequently start and stop, the measurement point data is temporarily stored, and the curve update of steps S3 to S5 is uniformly executed after the unit stops, and the update result is used as the benchmark for the next operating cycle; and it also includes a post-wash benchmark iteration step: after the unit completes offline washing and operates stably, the real-time characteristic curve family with qualified reliability is marked as the post-wash short-term benchmark, and the single-unit correction factor is calculated to offset the inherent offset caused by manufacturing tolerances, assembly deviations and irreversible degradation. The short-term benchmark replaces the benchmark curve family as the starting point for degradation correction within a preset validity period.