Data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units

Abstract

This invention discloses a data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units, relating to the field of data processing. It parses continuous time-series operating data of gas-fired steam combined cycle units into a sequence of thermodynamic vector lines, outputting symmetrical and asymmetrical operating segments. Using historical operating samples, it trains and calls a regression model based on asymmetrical load correction factors to identify and suppress distortions caused by abnormal steam metering and nonlinear mixing under asymmetrical heating conditions. It outputs equivalent steam flow sets and equivalent thermodynamic state sets, constructs a virtual equivalent energy channel oriented towards the shared header and heating extraction steam topology, calculates the overall plant thermal efficiency, generates an energy efficiency reliability characterization quantity, and generates pseudo-high energy efficiency alarms and fluctuation suppression display parameters based on the energy efficiency reliability characterization quantity. This ensures optimized load allocation and refined management of multi-shaft units under complex operating conditions.

Description

Technical Field

[0001] This invention relates to the field of data processing, specifically to a data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units. Background Technology

[0002] For gas turbine combined cycle units employing a "two-to-one" multi-shaft configuration (i.e., two gas turbines driving one steam turbine), during the winter heating season, they often need to operate under asymmetrical heating conditions with extremely unbalanced loads between the two units or with one unit operating and the other on standby. Under such conditions, the plant-wide thermal efficiency indicators displayed in the distributed control system (DCS) often exhibit falsely high energy efficiency readings that violate thermodynamic design limits. For example, the displayed efficiency may be higher than the rated design value, or it may exhibit non-physical, high-frequency, violent oscillations even when there is no sudden change in fuel quantity. This makes it impossible for operators to make effective combustion adjustments and load distribution based on this distorted energy efficiency data.

[0003] The root cause of the aforementioned calculation errors lies in the fact that existing energy efficiency analysis models are generally based on the idealized assumption of linear superposition of fluid networks, assuming that the steam generated by two waste heat boilers does not interfere with each other when it merges into a shared main steam header. However, under asymmetric loads, the high-momentum fluid on the strong side exerts significant aerodynamic compression and shearing on the low-momentum fluid on the weak side, causing distortion of the flow field structure at the confluence of the shared header. This renders the flow measurement model based on the standard differential pressure principle ineffective due to the change in the velocity profile, leading to severe distortion in the weighted calculation of steam flow rate and enthalpy. Traditional steady-state thermodynamic calculation methods lack a data correction mechanism for this nonlinear effect of fluid collisions, and cannot accurately decouple the actual energy efficiency contribution of each heat source in the digital domain, thus producing the aforementioned calculation fallacy of falsely high energy efficiency.

[0004] To address the aforementioned shortcomings, a technical solution is provided. Summary of the Invention

[0005] To address the technical problems mentioned in the background section, this invention is proposed. This invention provides a data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units.

[0006] This invention is achieved through the following technical solution: a data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units, the method comprising the following steps:

[0007] A data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units, comprising the following steps:

[0008] The continuous time-series operation data of the gas-fired steam combined cycle unit is parsed into a sequence of thermodynamic vector line segments. The sequence of thermodynamic vector line segments is then sieved based on the physical limit of thermal inertia to output symmetrical and asymmetrical operation segments.

[0009] Based on symmetrical and asymmetrical operating sections, the regression model of asymmetrical load correction factor is trained using historical operating samples and called to identify and suppress the distortion caused by abnormal steam metering and nonlinear mixing under asymmetrical heating conditions, and output equivalent steam flow set and equivalent thermodynamic state set.

[0010] Using the equivalent steam flow set and the equivalent thermodynamic state set as input, a virtual equivalent energy channel oriented towards the shared main pipe and heating extraction steam topology is constructed, the overall plant thermal efficiency is calculated, an energy efficiency reliability characterization quantity is generated, and a reliable energy efficiency index set is output.

[0011] The set of reliable energy efficiency indicators is written into the indicator point table of the distributed control system, and pseudo-high energy efficiency alarm and fluctuation suppression display parameters are generated based on the energy efficiency reliability characterization quantity.

[0012] Furthermore, the step of outputting the symmetrical and asymmetrical operating segments includes:

[0013] Continuous time-series operation data of the gas-fired steam combined cycle unit is acquired. The second derivative of the continuous time-series operation data is scanned to identify the physical inflection point of the load change rate. The physical inflection point is used as the feature anchor point, and the operation trajectory between two adjacent feature anchor points is discretized and decomposed to output a sequence of thermodynamic vector line segments containing heat absorption vector, steady-state vector and heat release vector.

[0014] Furthermore, the step of outputting the symmetrical and asymmetrical operating segments also includes:

[0015] Based on the thermal inertia physical limit of the unit's metal components, the minimum physical transition time of each segment in the thermal vector line segment sequence is calculated and compared with the geometric time to eliminate thermal transient segments.

[0016] Based on the compared heat absorption vector, steady-state vector, and heat release vector, the time-domain closure and dual-machine coupling are determined. The sequence with balanced dual-machine load and forming a closed thermodynamic cycle is marked as a symmetrical operating segment, while the sequence without forming a closed cycle, with residual thermal stress, or with unbalanced dual-machine load is marked as an asymmetrical operating segment.

[0017] Furthermore, the step of outputting the equivalent steam flow set and the equivalent thermodynamic state set includes:

[0018] A symmetrical operating section sample was selected as the benchmark sample to train a benchmark mapping model from the operating condition descriptor to steam metering and thermodynamic state, so as to obtain the statistical correlation before the failure of the linear superposition assumption of the shared main pipe.

[0019] Based on the benchmark mapping model, the benchmark prediction residual under the asymmetric operating section is calculated, and the physical characteristics of the flow field distortion at the confluence of the main pipes are coupled to construct a nonlinear compensation mechanism driven by physical causes, and generate an asymmetric load correction factor.

[0020] Furthermore, the step of generating the asymmetric load correction factor also includes:

[0021] Input the working condition descriptor of the asymmetric section into the benchmark mapping model, obtain the theoretical prediction value based on the symmetry assumption, and calculate the benchmark prediction residual between the theoretical prediction value and the actual effective measurement set.

[0022] Based on the spatial geometric topology of the junction of the outlets of the two waste heat boilers and the main pipe, and combined with the real-time flow rate and density of the two steam streams, the momentum flux ratio and shear layer distribution during fluid collisions are deduced, and a set of physical characteristics of flow field distortion is generated.

[0023] Using the physical feature set of flow field distortion as the independent variable and the baseline prediction residual as the dependent variable, an asymmetric load correction factor regression model is constructed to analyze the contribution weight of physical distortion to measurement deviation and output an asymmetric load correction factor that includes fluid dynamic constraints.

[0024] Furthermore, the step of generating the physical feature set of flow field distortion includes:

[0025] A three-dimensional fluid computational domain is established based on the spatial geometric topology of the junction of the outlets of the two waste heat boilers and the main pipe to obtain the real-time mass flow rate and density parameters of the two steam streams. Based on the principle of momentum conservation, the momentum flux ratio of the two steam streams at the junction node is calculated, and the aerodynamic stagnation interface surface formed by the collision of the two fluid streams is solved accordingly.

[0026] Based on the spatial distribution of the aerodynamic stagnation interface surface in the three-dimensional fluid computation domain, the effective flow cross section and effective hydraulic diameter of the weak-side fluid after being squeezed by the strong-side fluid are calculated.

[0027] Furthermore, the step of generating the physical feature set of flow field distortion also includes:

[0028] Using the aerodynamic stagnation interface surface as the initial boundary condition for shear mixing, and combining the Kelvin-Helmholtz instability characteristics induced by the momentum flux ratio, a spatial distribution model of the shear mixing layer extending along the parent tube axis is constructed.

[0029] The physical spatial coordinates of the temperature sensor on the main pipe are projected into the spatial distribution model of the shear mixing layer, and the geometric deviation vector between the sensor probe position and the true energy centroid of the fluid is calculated. Combining the correction effect of the effective hydraulic diameter on flow measurement and the correction effect of the geometric deviation vector on enthalpy measurement, the physical feature set of flow field distortion is generated.

[0030] Furthermore, the step of outputting the equivalent steam flow set and the equivalent thermodynamic state set also includes:

[0031] The asymmetric load correction factor is applied to the actual measured data of steam metering and thermodynamic state corresponding to the asymmetric operating section. An iterative reweighting strategy based on robust regression is used to suppress outliers, and the equivalent steam flow set and the equivalent thermodynamic state set are output.

[0032] Furthermore, the step of outputting the reliable energy efficiency index set includes:

[0033] Using the equivalent steam flow set and the equivalent thermodynamic state set as input, a virtual main pipe thermodynamic state is generated based on the shared main pipe topology, so that the contributions of the two waste heat boilers to the turbine steam inlet and heating extraction steam can be allocated in the digital domain.

[0034] Based on the thermal state of the virtual main pipe, the equivalent energy quantification results of the power generation output channel, the heat supply output channel and the unavailable loss channel are calculated to form an equivalent energy channel set.

[0035] The overall plant thermal efficiency is calculated from the equivalent energy channel set, and the energy efficiency confidence characterization quantity is generated based on the reconstructed confidence interval, cross-channel consistency verification results and segment stability, and the confidence energy efficiency index set is output.

[0036] Furthermore, the step of generating pseudo-high energy efficiency alarm and fluctuation suppression display parameters includes:

[0037] The set of reliable energy efficiency indicators is mapped to indicator points that can be subscribed to by the distributed control system, and a quality label corresponding to the calculation cycle is output.

[0038] Based on the energy efficiency credibility representation quantity and the preset pseudo-high energy efficiency rule base, it is determined whether the current energy efficiency display is falsely high;

[0039] When a false high energy efficiency is detected, an alarm signal and visual suppression parameters are generated, and operational recommendations for asymmetric heating regulation are output for operators' reference.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] This invention overcomes the limitations of the idealized assumption of linear superposition of fluids in a shared main pipe in traditional energy efficiency calculations by constructing a physical model of flow field distortion based on momentum flux ratio and aerodynamic stagnation interface characteristics. By introducing an asymmetric load correction factor regression model, this method can quantify the aerodynamic compression and shearing effects of the high-momentum fluid on the strong side on the weak side, accurately identify and compensate for the measurement distortion caused by the narrowing of the effective flow cross section and the shift of the energy centroid. Thus, without increasing the number of hardware measurement points, it solves the problem of steam flow rate and enthalpy calculation deviation caused by changes in the flow field structure under asymmetric operation conditions in a two-to-one turbine unit.

[0042] This invention establishes a virtual equivalent energy channel and multi-dimensional energy efficiency reliability evaluation for shared topologies, enabling precise decoupling and tracing of the energy efficiency of independent heat sources in two-to-one units under asymmetric heating conditions. By integrating the logic of reconstructing confidence intervals, cross-channel consistency verification, and physical limit constraints to identify pseudo-high energy efficiency, this method can effectively eliminate "false high energy efficiency" readings that violate thermodynamic principles and non-physical data oscillations. It provides operators with energy efficiency indicators that have been corrected by fluid mechanics and have self-checking capabilities, ensuring optimized load allocation and refined management of multi-shaft units under complex variable operating conditions. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not drawn to scale according to the actual size, but are intended to show the main idea of ​​the present invention.

[0044] Figure 1 The flowchart shows the method for intelligent energy efficiency analysis of gas-fired steam turbine units based on data-driven approaches.

[0045] Figure 2 A schematic diagram showing the formation of the momentum flux ratio and aerodynamic stagnation interface surface in a three-dimensional fluid computational domain;

[0046] Figure 3 This is a schematic diagram showing the cross-sectional projection of the main pipe, the effective flow section, and the wetted perimeter calculation.

[0047] Figure 4 This is a schematic diagram showing the division of symmetrical and asymmetrical operating sections. Detailed Implementation

[0048] The technical solutions in 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 also within the scope of protection of the present invention.

[0049] In this embodiment, the gas-steam combined cycle unit is specifically applied to a two-to-one multi-shaft configuration architecture. This architecture refers to a thermodynamic cycle system composed of two gas turbines, two waste heat boilers, and one steam turbine. Each gas turbine is connected to one waste heat boiler, and the high-temperature, high-pressure steam generated by the two waste heat boilers merges into the same shared header through their respective branch pipes, thereby driving the same steam turbine to generate electricity. While this two-to-one structure improves system flexibility and efficiency under partial load, the physical topology of its shared header also introduces complex fluid dynamic coupling problems: when the two gas turbines are operating asymmetrically (e.g., uneven load, asynchronous start-stop, or differences in environmental boundary conditions), the two steam streams experience violent momentum collisions and shear compression at the header junction, causing traditional single-point flow and temperature measurements to deviate significantly from the true values, thus affecting the accuracy of the overall plant energy efficiency assessment.

[0050] Based on the above, such as Figure 1 As shown, the data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units includes the following steps:

[0051] like Figure 4 As shown, step S10: The continuous time-series operation data of the gas-steam combined cycle unit is parsed into a sequence of thermodynamic vector line segments. The thermodynamic vector line segment sequence is screened based on the physical limit of thermal inertia, and symmetrical operation segments and asymmetrical operation segments are output.

[0052] Step S10 includes the following sub-steps:

[0053] Step S11: Obtain continuous time-series operation data of the gas-fired steam combined cycle unit, perform second-order derivative scanning on the continuous time-series operation data to identify the physical inflection point of the load change rate, use the physical inflection point as the feature anchor point, and discretize and decompose the operation trajectory between two adjacent feature anchor points to output a sequence of thermodynamic vector line segments containing heat absorption vector, steady-state vector and heat release vector.

[0054] Considering the structural characteristics of a combined cycle gas turbine unit employing a two-to-one multi-shaft configuration (i.e., the first and second gas turbines jointly drive one steam turbine), the acquired continuous time-series operating data includes active power time-series data from both the first and second gas turbines. Since the two gas turbines may operate asynchronously with uneven load distribution or inconsistent start-stop states during actual operation, a dual-channel parallel processing mechanism is adopted in this implementation: a first data processing channel and a second data processing channel are established to independently acquire and calculate data from both gas turbines. The acquisition process is conducted through a distributed control system at a preset high-frequency sampling rate (e.g., 1 second / sample), and Savitzky-Golay digital filters are used to smooth the two original power sequences to eliminate high-frequency noise generated by electromagnetic interference and retain the true load change trend. The physical significance of this step is that by decoupling the independent heat source data of the multi-shaft unit, the interference of turbine power lag and multi-shaft superposition effect is eliminated, ensuring that the subsequently generated thermodynamic vector line segment sequence can accurately correspond to the original driving force of each combustion side, providing a data basis for analyzing the independent thermal inertia of a single gas turbine.

[0055] The implementation method for identifying physical inflection points of load change rate by scanning the second derivative of continuous time-series operating data is as follows: Numerical differentiation calculations are performed on the smoothed active power sequences of the first and second gas turbines, respectively, to obtain the corresponding first and second derivative sequences. Physical inflection points are defined as slope change inflection points and steady-state entry points. The determination logic for slope change inflection points is: when the value of the second derivative sequence crosses zero, and the absolute value of the first derivative at the corresponding moment is greater than a preset fluctuation threshold, that moment is determined to be a slope change inflection point; the determination logic for steady-state entry points is: when the absolute value of the first derivative falls from greater than the fluctuation threshold to less than the quiescent threshold, that moment is determined to be a steady-state entry point. All identified slope change inflection points and steady-state entry points are arranged in chronological order to form the first feature anchor set and the second feature anchor set, respectively. The technical purpose of physical inflection point identification is to capture the critical moment of qualitative change in the operating state of the thermodynamic system using the mathematical derivative extreme value characteristics, and to segment the continuous analog quantity curve into discrete event nodes with clear thermodynamic meaning. The fluctuation threshold and the quiet threshold are obtained by selecting long-term stable operating data of the unit under historical rated operating conditions, calculating the standard deviation of the power fluctuation rate during this period, setting the fluctuation threshold to 3 times the standard deviation, and setting the quiet threshold to 1 times the standard deviation.

[0056] The specific implementation method for constructing a thermal vector line segment sequence based on physical inflection points is as follows: Traverse the first feature anchor point set, using two adjacent feature anchor points as the start and end points, extract data segments corresponding to the time period, and construct directed line segments connecting the start and end point coordinates to form the first thermal vector line segment sequence; similarly, traverse the second feature anchor point set to form the second thermal vector line segment sequence. For each thermal vector line segment in these two sequences, based on its slope in the load-time coordinate system... Perform attribute classification (where Representing the (Sequence number of the line segment). Attribute classification is based on positive and negative judgment thresholds: if the first... The slope of the line segment If the value is greater than the positive judgment threshold, the line segment is marked as an endothermic vector; if the first... The slope of the line segment If the value is less than the negative threshold, the line segment is marked as a heat-releasing vector; if the first... The slope of the line segment Between the negative and positive decision thresholds, this line segment is marked as a steady-state vector. The physical mechanism of attribute classification is as follows: When the slope is positive, the gas turbine load increases, the turbine exhaust temperature rises, the fluid temperature is higher than the metal wall temperature, and the metal components absorb heat from the fluid to increase their internal energy, thus being in a heat storage state, hence defined as an endothermic vector; when the slope is negative, the gas turbine load decreases, the fluid temperature decreases, the metal wall temperature is higher than the fluid temperature, and the metal components release sensible heat to the fluid, thus being in a heat release state, hence defined as a heat release vector; when the slope approaches zero, the fluid and the metal wall reach dynamic thermal equilibrium, with no heat storage or heat release phenomenon, hence defined as a steady-state vector. The above-mentioned positive and negative judgment thresholds are obtained as follows: The technical specifications of the gas turbine control system are consulted to obtain the rated active power value of the unit, and the dead zone ratio coefficient is determined. The dead zone ratio coefficient is determined by: obtaining continuous active power time-series data of the unit during historical steady-state grid-connected operation without external frequency regulation commands, and calculating the standard deviation of this continuous active power time-series data; dividing three times the standard deviation by the rated active power of the unit, and determining the quotient as the dead zone ratio coefficient. The rated active power value is multiplied by the dead zone ratio coefficient to calculate the power control dead zone value. The physical meaning of the power control dead zone value represents the allowable natural fluctuation range of the control system. A minimum effective time span is set. The minimum effective time span is obtained by: extracting the data sampling period of the distributed control system and obtaining the step response delay time of the unit's underlying thermal sensors. The steps for determining the step response delay time are as follows: From the unit's historical operating data, extract the timestamp of the step change in the fuel control valve opening command as the first characteristic moment; extract the temperature sequence measured by the corresponding turbine exhaust temperature sensor, and use a sliding window algorithm to identify the first time point where the first derivative of the temperature sequence exceeds a preset silent threshold as the second characteristic moment; calculate the time difference between the second characteristic moment and the first characteristic moment, and determine this time difference as the step response delay time of the unit's underlying thermal sensor. Add the data sampling period to the step response delay time, and determine the sum as the minimum effective time span, for example, one minute. Divide the calculated power control dead zone value by the minimum effective time span to obtain a positive value as a positive judgment threshold, which represents the positive power change rate exceeding the control dead zone per unit time; divide the negative of the calculated power control dead zone value by the minimum effective time span to obtain a negative value as a negative judgment threshold, which represents the negative power change rate exceeding the control dead zone per unit time. The physical meaning of setting a negative judgment threshold is that only when the power change rate exceeds the inherent dead zone of the control system is it considered an active load change driven by the thermodynamic cycle, thereby causing significant heat absorption or release behavior of the metal components; while small fluctuations within the threshold range are considered as the system's own regulatory oscillations, which do not produce macroscopic thermal inertia effects.

[0057] Through the above steps, two independent sequences of thermodynamic vector segments are output, which completely characterize the dynamic thermodynamic processes of the first and second gas turbines, respectively. The heat absorption vector corresponds to the heat storage process when the gas turbine load increases and the metal components absorb heat from the fluid; the heat release vector corresponds to the heat release process when the gas turbine load decreases and the metal components release heat to the fluid; the steady-state vector corresponds to the process when the fluid and the metal wall reach dynamic thermal equilibrium. This method, based on dual-channel parallel processing and slope feature classification, can transform the complex and variable operating curves of the two-to-one turbine unit into a structured vector sequence, accurately defining the heat transfer direction and intensity of each gas turbine in different time periods. This provides a computational unit with physical properties for the subsequent calculation of the thermal inertia correction coefficient of each gas turbine, ensuring the calculation accuracy and logical completeness of the multi-shaft unit energy efficiency assessment model.

[0058] Step S12: Based on the thermal inertia physical limit of the unit's metal components, calculate the minimum physical transition time of each segment in the thermal vector line segment sequence, and compare it with the geometric time to eliminate thermal transient segments; based on the compared heat absorption vector, steady-state vector, and heat release vector, determine the time domain closure and dual-machine coupling, mark the sequence with balanced dual-machine load and forming a closed thermal cycle as a symmetrical operating segment, and mark the sequence without forming a closed cycle, with residual thermal stress, or dual-machine load imbalance as an asymmetrical operating segment.

[0059] The implementation method for calculating the minimum physical transition time based on the thermal inertia physical limit of the unit's metal components is as follows: Multiply the total mass value of the gas turbine blades and combustion chamber metal components by their specific heat capacity to calculate the total heat that the combustion chamber metal components must absorb or release to raise or lower the temperature by one unit, i.e., the heat capacity. Based on the allowable thermal stress intensity threshold of the materials used in the combustion chamber metal components under high-temperature conditions as specified in the mechanics of metal materials handbook, determine the maximum heat flux density and corresponding maximum heat exchange power allowed to pass through the surface of the combustion chamber metal components per unit time, ensuring no thermal fatigue cracks occur. Divide the maximum heat exchange power by the heat capacity; the quotient is the maximum temperature change amplitude that the combustion chamber metal components can physically withstand per unit time. This maximum temperature change amplitude is defined as the limiting heating rate and the limiting cooling rate, respectively, according to their positive and negative directions. For each segment in the thermodynamic vector line segment sequence, extract the corresponding power change amplitude. Consult the load and exhaust temperature characteristic curves provided by the gas turbine manufacturer to map and convert the power change amplitude into the equivalent temperature change amplitude on the turbine exhaust side. Dividing the equivalent temperature change amplitude by the corresponding limiting heating rate or limiting cooling rate, the theoretical shortest time required for the line segment to complete the temperature change under physical heat conduction constraints is calculated, i.e., the minimum physical transition time. The projected length of the thermodynamic vector line segment on the time axis is calculated, i.e., the time difference between the endpoint timestamp and the starting timestamp, defined as the geometric duration. The minimum physical transition time is compared with the geometric duration: if the geometric duration is less than the minimum physical transition time, it indicates that the load change rate represented by the line segment exceeds the physical limit of the metal material's heat conduction, belonging to non-thermodynamically dominated transient fluctuations or high-frequency noise on the electrical side; the line segment is marked as a thermal transient line segment and discarded. If the geometric duration is greater than or equal to the minimum physical transition time, the line segment is retained. The physical significance of this step is: using the inherent heat capacity properties and heat transfer rate constraints of the metal material as a physical filter, identifying and discarding pseudo-thermal signals that fluctuate in data values ​​but cannot cause substantial changes in the metal temperature field in physical reality, ensuring that the retained line segments truly reflect the effective dynamic process constrained by the unit's thermal inertia.

[0060] The implementation method for determining time-domain closure based on the compared endothermic vector, steady-state vector, and exothermic vector is as follows: In the first and second thermodynamic vector segment sequences after removing transient thermal segments, candidate segments that conform to the topological structure of steady-state vector initiation, endothermic or exothermic vector as process, and steady-state vector termination are searched using a sliding window. For each candidate segment, the average power value of the initiating steady-state vector and the average power value of the terminating steady-state vector are extracted. A threshold for residual thermal stress is set. The threshold is obtained by selecting the power fluctuation range when the unit reaches its first thermal equilibrium after a cold start, and taking 110% of the maximum value of this fluctuation range as the threshold. The physical basis for setting it to 110% is that, in the actual operation of the unit under reset state, in addition to pure thermal stress, there is also about 5% of sensor measurement uncertainty and background noise caused by environmental heat dissipation fluctuations. The 10% margin can both accommodate these natural fluctuations caused by non-stress and strictly intercept any substantial unreleased thermal gradient state. If this value is too large, it may mask the early local heat accumulation phenomenon inside the metal, causing the time domain closure judgment to fail. The absolute value of the difference between the average power value of the initial steady-state vector and the average power value of the final steady-state vector is calculated. If the absolute value of this difference is less than the residual thermal stress determination threshold, it is determined that the segment constitutes a closed heating-to-cooling cycle or a cooling-to-heating cycle in the time domain. The metal component has undergone a complete thermal expansion and contraction process and returned to the initial thermal equilibrium state, with no residual thermal stress. If the absolute value of this difference is greater than the residual thermal stress determination threshold, it is determined that the segment has not formed a closed cycle, and unreleased thermal gradients or thermal stresses still remain inside the metal component. The technical purpose of the time-domain closure determination is to screen out operating segments where the unit's thermal state is completely reset, eliminating the benchmark drift error caused to the energy efficiency regression model by the inconsistency between the enthalpy of the initial and final states.

[0061] The implementation method for determining the coupling relationship between the two gas turbines and marking the final segment is as follows: Align the first and second thermal vector line segment sequences on the same time axis. For any time slice, calculate the absolute value of the difference between the power of the first and second gas turbines, and then divide it by the arithmetic mean of the power of the two gas turbines to obtain the load deviation. Set a load balance threshold for the two turbines. The load balance threshold is obtained by: extracting historical sample sets of parallel operation of the two turbines from the historical operation database, calculating the absolute value of the difference in superheater outlet steam temperature between the two waste heat boilers in each historical sample; selecting a stable subset whose absolute value is less than or equal to 3 degrees Celsius (this temperature difference is the physical observation critical value calibrated by the monitoring system that triggers a sudden change in stress on one side of the pipe wall); extracting all the load deviation data of the two turbines in this stable subset, performing statistical distribution fitting on them, and extracting the load deviation value corresponding to the cumulative probability density function reaching 95%, defining the load deviation value as the load balance threshold for the two turbines. If, within a certain time period, the thermodynamic vector line segment sequences of both the first and second gas turbines satisfy the aforementioned time-domain closure determination conditions, and the load deviation at all sampling points within that time period is less than the dual-turbine load balance threshold, then that time period is determined to be a dual-turbine load balance and constitutes a closed thermodynamic cycle, and is marked as a symmetrical operating segment. Conversely, if, within a certain time period, the sequence of any gas turbine does not form a closed cycle, has residual thermal stress, or the load deviation exceeds the dual-turbine load balance threshold leading to dual-turbine load imbalance, then that time period is marked as an asymmetrical operating segment. The physical significance of the dual-turbine coupling determination lies in ensuring that the data selected for the symmetrical operating segment not only satisfies thermal balance at the single-turbine level but also satisfies fluid dynamic symmetry at the dual-turbine level. This eliminates the nonlinear mixing, backflow, or compression effects of the waste heat boiler outlet steam within the shared header caused by excessive load differences between the two gas turbines, thereby providing idealized physical boundary conditions for subsequently establishing a pure benchmark energy efficiency model.

[0062] By employing transient elimination based on thermal inertia limits, time-domain thermal closure verification, and dual-machine load balancing screening, high-quality steady-state symmetrical samples can be accurately extracted from historical operating data. This approach effectively solves the problem of energy efficiency calculation distortion caused by metal heat storage release and dual-machine flow field interference in complex variable operating conditions of gas-fired combined cycle units. By strictly defining symmetrical and asymmetrical sections, a solid data classification foundation is laid for the subsequent construction of benchmark and modified models, improving the robustness and reliability of energy efficiency analysis results under different operating scenarios.

[0063] Step S20: Based on the symmetrical and asymmetrical operating sections, the asymmetrical load correction factor regression model is trained using historical operating samples and called to identify and suppress the distortion caused by abnormal steam metering and nonlinear mixing under asymmetrical heating conditions, and output the equivalent steam flow set and the equivalent thermodynamic state set.

[0064] Step S20 includes the following sub-steps:

[0065] Step S21: Select the symmetrical operating section sample as the benchmark sample, and train the benchmark mapping model from the operating condition descriptor to steam metering and thermodynamic state to obtain the statistical correlation before the failure of the shared main pipe linear superposition assumption.

[0066] A symmetrical operating segment sample is selected as the benchmark sample to construct a training set: all time series segments marked as symmetrical operating segments are extracted from the output of step S12. These segments are cleaned to remove null values ​​and bad points caused by sensor failures or communication packet loss. The remaining valid data points are aligned by timestamps to form the benchmark sample set. For each sampling moment in the benchmark sample set, a condition description sub-vector and a steam metering thermodynamic state vector are constructed. The condition description sub-vector contains key physical quantities that determine the exhaust energy of the gas turbine, specifically including: the active power of the first gas turbine, the active power of the second gas turbine, the inlet guide vane opening of the first gas turbine, the inlet guide vane opening of the second gas turbine, ambient temperature, ambient atmospheric pressure, and ambient relative humidity. The steam metering thermodynamic state vector contains response physical quantities on the waste heat boiler side and the main pipe side, specifically including: the main steam flow rate of the first waste heat boiler, the main steam flow rate of the second waste heat boiler, the main steam temperature of the first waste heat boiler, the main steam temperature of the second waste heat boiler, and the combined pressure of the shared main pipe. The technical purpose of selecting the symmetrical operating section as the benchmark sample is that, under the symmetrical operating condition of dual-unit load balance, the steam velocity, momentum and pressure distribution at the outlet of the two waste heat boilers are basically the same, and the fluid mixing process in the shared header is in an ideal collision or parallel confluence state in fluid mechanics. There are no significant flow deviation, backflow or shear compression phenomena. The operating data at this time is closest to the ideal physical model of linear superposition of fluid networks, and can truly reflect the background energy efficiency characteristics of the unit under the condition of no flow field distortion interference.

[0067] The implementation method for training the benchmark mapping model from operating condition descriptors to steam metering and thermodynamic state is as follows: a nonlinear mapping model is constructed using a Gaussian process regression algorithm. Specifically, the operating condition descriptor vector and the steam metering thermodynamic state vector in the benchmark sample set are Z-score standardized, i.e., the mean is subtracted and the standard deviation is divided to eliminate the interference of different physical dimensions on the model weights. A radial basis function is selected as the kernel function to map the low-dimensional operating condition descriptor to a high-dimensional feature space to capture the complex nonlinear thermodynamic relationship between gas turbine load and steam parameters. The standardized operating condition descriptor vector is used as the model input, and the standardized steam metering thermodynamic state vector is used as the model output. The regression coefficient matrix is ​​solved by minimizing the sum of squared prediction errors and the sum of regularization terms. During training, the bandwidth parameter and regularization coefficient of the kernel function are determined using K-fold cross-validation. The benchmark sample set is randomly divided into K subsets, where K is 5 or 10. One subset is selected alternately as the validation set, and the rest are used as the training set. The parameter combination that minimizes the mean square error of the validation set is selected as the optimal model parameters. The trained benchmark mapping model characterizes the standard statistical correlation between input conditions and output thermodynamic states under ideal symmetric flow field conditions. The physical significance of this step lies in using a data-driven approach to solidify the unit's input-output response function under "flow field health" conditions. This is equivalent to establishing a virtual standard unit in digital space, unaffected by asymmetric flow field distortion, providing a unique benchmark for subsequent quantification of the degree of flow field distortion under asymmetric conditions.

[0068] The implementation method for obtaining the statistical correlation before the failure of the linear superposition assumption for the shared header is as follows: The trained benchmark mapping model is saved, and the weight matrix and kernel function structure within the benchmark mapping model represent the statistical correlation. Here, the linear superposition assumption means that, ideally, the total steam flow from the waste heat boilers corresponding to the two gas turbines should be equal to the algebraic sum of the flow rates of the two individual turbines when they converge into the header, and they should not interfere with each other. However, in actual asymmetric operation, the stronger fluid will compress the weaker fluid, causing distortion in the flow meter readings, i.e., the linear superposition assumption fails. By training the model using only data from the symmetric section, the model learns the physical laws when the linear superposition assumption has not yet failed, or when the degree of failure can be considered as a systematic error and ignored. Therefore, when the benchmark mapping model is subsequently called, for any input operating condition descriptor, its output value represents the theoretical steam parameters assuming no distortion of the flow field.

[0069] By constructing and training a benchmark mapping model based on symmetric samples, high-dimensional nonlinear regression technology was used to accurately capture the thermodynamic response characteristics of a gas-fired steam combined cycle unit under ideal flow field conditions. This approach cleverly decouples complex fluid dynamic disturbances from the unit's inherent thermodynamic conversion performance, establishing a mathematical benchmark capable of quantifying and evaluating normal operating conditions. When asymmetric operating condition data is subsequently input into the model, the deviation between the model's predicted values ​​and the actual measured values ​​no longer includes the unit's performance degradation or environmental impacts, but purely reflects the flow field distortion and metering inaccuracies caused by the load imbalance between the two units. This provides a high-confidence residual signal for calculating the asymmetric load correction factor, ensuring the benchmark consistency and logical rigor of the energy efficiency analysis method across the entire operating range.

[0070] Step S22: Calculate the baseline prediction residual under the asymmetric operating section based on the baseline mapping model, and couple the physical characteristics of the flow field distortion at the confluence of the main pipes to construct a nonlinear compensation mechanism driven by physical causes and generate an asymmetric load correction factor.

[0071] Step S221: Input the working condition descriptor of the asymmetric section into the benchmark mapping model, obtain the theoretical prediction value based on the symmetry assumption, and calculate the benchmark prediction residual between the theoretical prediction value and the actual effective measurement set.

[0072] The implementation method for obtaining theoretical predictions based on the symmetry assumption is as follows: The benchmark mapping model trained in step S21 is invoked, and the mean vector and standard deviation vector used during model training are read. Operating condition descriptor data within the asymmetric operating section are extracted, and Z-score standardization is performed using the mean vector and standard deviation vector to ensure that its feature distribution is consistent with the training space. The standardized operating condition descriptor is input into the benchmark mapping model for forward inference, outputting a standardized prediction result vector. Using the mean and standard deviation corresponding to the steam metering thermodynamic state vector in step S21, the prediction result vector is denormalized to obtain a numerical sequence with actual physical dimensions, i.e., the theoretical prediction value based on the symmetry assumption. The energy conversion law learned under an ideal symmetric flow field is used to virtually reconstruct the current asymmetric operating condition. Since the benchmark mapping model has not learned flow field distortion characteristics, the output value represents the ideal flow rate and thermodynamic state that should be presented when the steam discharged from the two gas turbines does not interfere with each other at the junction of the main pipe and strictly follows the linear superposition principle, thus constructing a theoretical reference system for measuring the degree of flow field distortion.

[0073] The implementation method for constructing the actual effective measurement set and calculating the benchmark prediction residual is as follows: Extract measurement point data related to the steam metering thermodynamic state from the original collected data of the asymmetric operating section. Verify the validity of this measurement point data by using the Laida criterion to remove outliers: Calculate the arithmetic mean and standard deviation of the data within the asymmetric operating section. If the absolute value of the difference between the measured value and the mean at a certain moment is greater than three times the standard deviation, the point is determined to be an invalid measurement and is removed, or it is supplemented using nearest neighbor interpolation. The processed data sequence constitutes the actual effective measurement set. Subtract the corresponding theoretical prediction value based on the symmetry assumption from the values ​​in the actual effective measurement set at the same moment to obtain a multi-dimensional vector sequence containing flow deviation, temperature deviation, and pressure deviation. Extract the flow deviation component from this vector sequence and divide the flow deviation component by the flow value in the corresponding theoretical prediction value based on the symmetry assumption to obtain the instantaneous relative error sequence. Calculate the arithmetic mean of this instantaneous relative error sequence to obtain a scalar value reflecting the overall metering deviation level of the section, which is defined as the benchmark prediction residual. The physical meaning of the baseline prediction residual is that it quantifies the average degree of difference between the actual physical process and the ideal linear superposition model. Under asymmetric conditions, the compression and shearing of the strong-side fluid on the weak-side fluid, as well as the formation of local vortices, cause distortion in the velocity distribution at the flowmeter cross-section, resulting in sensor readings continuously deviating from the theoretical true value derived from the law of conservation of energy. The baseline prediction residual aggregates this measurement distortion caused by hydrodynamic effects and the deviation of thermodynamic parameters caused by nonlinear mixing into a statistically significant relative error value, providing a unique quantitative target for constructing the nonlinear compensation model and calculating the correction factor in subsequent step S223.

[0074] The complex asymmetric flow field disturbance problem is transformed into a residual analysis problem between model predictions and actual observations. This approach uses a benchmark mapping model as a filter to eliminate the conventional influence of gas turbine load variations on steam parameters, ensuring that the remaining residual signal purely reflects the nonlinear effects of the main pipe flow field caused by the load imbalance between the two turbines. This provides a precise error signal source for subsequent steps to establish a physically driven nonlinear compensation mechanism, ensuring that the correction factor compensates for flow field distortions rather than incorrectly correcting normal performance fluctuations of the unit.

[0075] Step S222: Based on the spatial geometric topology of the junction of the outlets of the two waste heat boilers and the main pipe, and combined with the real-time flow rate and density of the two steam streams, the momentum flux ratio and shear layer distribution during fluid collision are deduced to generate a set of physical characteristics of flow field distortion.

[0076] Step S2221: Establish a three-dimensional fluid computation domain based on the spatial geometric topology of the junction of the outlets of the two waste heat boilers and the main pipe, and obtain the real-time mass flow rate and density parameters of the two steam streams; based on the principle of momentum conservation, calculate the momentum flux ratio of the two steam streams at the junction node, and solve the aerodynamic stagnation interface surface formed by the collision of the two fluid streams accordingly.

[0077] Based on the spatial distribution of the aerodynamic stagnation interface surface in the three-dimensional fluid computation domain, the effective flow cross section and effective hydraulic diameter of the weak-side fluid after being squeezed by the strong-side fluid are calculated.

[0078] like Figure 2 The figure shows a schematic diagram of the momentum flux ratio and the formation of the aerodynamic stagnation interface surface in the three-dimensional fluid computation domain.

[0079] Reference Figure 2 The diagram illustrates a three-dimensional Cartesian fluid computation coordinate system. The origin is set at the geometric center where the centerlines of the two branch pipes intersect with the centerline of the main pipe. The positive Z-axis points downstream along the fluid flow direction in the main pipe, the positive Y-axis is perpendicular to the branch pipe plane and points upwards, and the positive X-axis points from the weaker fluid to the stronger fluid within the branch pipe plane. Within this coordinate system, a mesh is created to define spatial coordinate nodes, forming a three-dimensional fluid computation domain encompassing the inlets of the two branch pipes and the mixing section of the main pipe. The three-dimensional surface with shaded mesh in the diagram represents the aerodynamic stagnation interface surface. This surface is not formed by a solid partition, but rather by the momentum conflict between the stronger and weaker fluids within a confined space, creating an invisible aerodynamic barrier. This barrier visually demonstrates how the stronger fluid, through inertia, compresses the weaker fluid, preventing it from filling the physical pipe and forcing it to flow within a compressed spatial region, thus defining the actual flow boundaries of the two fluids. The section AA shown by the dashed line in the figure is used to indicate the spatial location for axial cross-sectional sectioning of the main pipe mixing section. It serves as the cross-sectional reference for converting the three-dimensional flow field distortion characteristics into a two-dimensional projection for geometric quantification calculations. See details... Figure 3 .

[0080] Combination Figure 2The diagram shows the momentum flux ratio and the formation of the aerodynamic stagnation interface surface in the three-dimensional fluid computational domain. A three-dimensional Cartesian fluid computational coordinate system is established in the region where the two steam lines converge. The geometric center point where the center lines of the two branch lines intersect with the center line of the main line is set as the origin. The direction of fluid flow along the main line and pointing downstream is set as the positive Z-axis. The direction perpendicular to the plane where the two branch lines are located and pointing upwards is set as the positive Y-axis. The direction perpendicular to the Z-axis and pointing from the weak side fluid to the strong side fluid in the plane where the two branch lines are located is set as the positive X-axis. Based on the spatial geometric topology of the junction of the outlets of the two waste heat boilers and the main line, a three-dimensional fluid computational domain including the inlets of the two branch lines and the mixing section of the main line is established in this coordinate system. The computational domain is meshed to determine the spatial coordinate nodes.

[0081] The average velocity of the two steam streams is calculated based on the real-time monitored mass flow rate and fluid density. The average velocity of the strong-side fluid is obtained by dividing the strong-side fluid mass flow rate by the product of the strong-side fluid density and the cross-sectional area of ​​the strong-side branch pipe. The average velocity of the weak-side fluid is obtained by dividing the weak-side fluid mass flow rate by the product of the weak-side fluid density and the cross-sectional area of ​​the weak-side branch pipe. Based on this, the momentum flux ratio is defined. The momentum flux ratio is physically defined as the ratio of the momentum flux of the strong-side fluid to the momentum flux of the weak-side fluid. Numerically, it is equal to the product of the density of the strong-side fluid and the square of the average velocity of the strong-side fluid divided by the product of the density of the weak-side fluid and the square of the average velocity of the weak-side fluid.

[0082] like Figure 2 The dashed surface shown represents the aerodynamic stagnation interface, a critical surface where two fluids are immiscible and in pressure equilibrium. To accurately describe this surface, an equilibrium equation for the aerodynamic stagnation interface considering the momentum flux ratio decay characteristics is constructed. This equation is based on the following: within a confined pipe space, the high-momentum fluid exerts a squeezing effect on the low-momentum fluid. This squeezing effect is not uniformly distributed but exhibits a specific decay law as the fluid diffuses within the pipe. Based on free jet theory and confined space mixing layer theory, the core momentum of the stronger fluid dissipates along the Z-axis and radially along the X and Y axes. Therefore, the interface location depends on the local dynamic pressure equilibrium. The aerodynamic stagnation interface equilibrium equation is as follows:

[0083]

[0084] In the equation, the left-hand side term represents the strong-side fluid in spatial coordinates. The effective impact dynamic pressure generated at the point is represented by the right-hand side of the equation, which represents the resistance dynamic pressure generated by the fluid on the weak side at the same point. The theoretical dynamic pressure representing the strong-side fluid is half the density of the strong-side fluid multiplied by the square of the strong-side fluid velocity. Representing the weak-side fluid in spatial coordinates The local dynamic pressure at a certain point is taken in engineering calculations as half the density of the weak-side fluid multiplied by the square of the velocity of the weak-side fluid. This is the momentum decay coefficient function, constructed based on the self-similarity theory of turbulent jets. The function value decays in a Gaussian distribution from the strong-side fluid central axis outwards into the surrounding space, and the characteristic width of the decay is related to the momentum flux ratio. The square root of the numbers is positively correlated, that is... The larger the value, the stronger the momentum retention capacity of the strong-side fluid in space, and the slower the decay coefficient decreases with distance; conversely, the smaller the value, the faster the decay. The specific function type can be a Gaussian function or an exponential decay function. This embodiment does not impose strict constraints on the specific mathematical form of the function, as long as it can characterize the physical law that momentum decays with increasing spatial distance and the decay rate is controlled by the momentum flux ratio.

[0085] Since the equilibrium equations are nonlinear, an iterative solution process is used to determine the aerodynamic stagnation interface: An initial plane is assumed as the aerodynamic stagnation interface; coordinates are substituted into the interface grid nodes to calculate the difference between the effective impact dynamic pressure on the strong side (left side of the equation) and the resistive dynamic pressure on the weak side (right side of the equation); the position is corrected based on the difference. If the effective impact dynamic pressure on the strong side is greater than the resistive dynamic pressure on the weak side at a certain point, the point is moved along the normal direction towards the weak side region; otherwise, it is moved towards the strong side region. The step size is proportional to the pressure difference. The above steps are repeated until the pressure difference at all interface nodes is less than a preset convergence threshold. The surface determined at this point is the final aerodynamic stagnation interface surface. This aerodynamic stagnation interface surface... Figure 2 The diagram visually illustrates its physical significance: in the fluid confluence region without a physical partition, the high-kinetic-energy fluid compresses the invisible aerodynamic barrier formed by the low-kinetic-energy fluid through inertia. The invisible aerodynamic barrier defines the actual space that each of the two fluids can occupy, and is a geometric expression of the fluid collision effect.

[0086] like Figure 3 The diagram shown is a schematic representation of the cross-sectional projection of the main pipe, the effective flow section, and the wetted perimeter calculation.

[0087] Figure 3 This is a key step in step S2221 of the present invention, which is used to transform the complex physical phenomenon of three-dimensional fluid collision into two-dimensional geometric parameters for quantitative calculation. Figure 3 This image shows a cross-section of a shared header in a gas-fired combined cycle unit. Inside the circular physical pipe section, the following key geometric elements are included: the physical inner wall of the header: i.e., the original metal inner wall boundary of the pipe; and the projection curve of the aerodynamic stagnation interface: representing the preceding steps (…). Figure 2The projection of the three-dimensional aerodynamic stagnation interface surface calculated in the model onto the two-dimensional cross-section. The aerodynamic stagnation interface projection curve divides the main pipe cross-section into two regions. The weak-side fluid region (shaded area): a closed region enclosed by the physical inner wall of the main pipe and the aerodynamic stagnation interface projection curve. This closed region represents the actual flow space that the weak-side fluid can occupy under the compression of the strong-side fluid. Based on... Figure 3 The geometric figure shown in the invention uses numerical integration to calculate the following two fluid dynamic parameters: effective flow cross section: i.e. Figure 3 The area marked as shaded is not the entire area of ​​the circular tube, but rather the area remaining after the stronger fluid has occupied it. Wet perimeter: For fluids that do not completely fill the circular tube, the wet perimeter consists of two boundary lengths: Physical wall contact boundary: the arc length of the contact between the weaker fluid and the physical inner wall of the main tube. Aerodynamic compression boundary: the length of the projected line segment of the contact between the weaker fluid and the aerodynamic stagnation interface. These two lengths are obtained by summing them using line integrals.

[0088] refer to Figure 3 The diagram showing the cross-sectional projection of the main pipe, the effective flow section, and the wetted perimeter calculation illustrates how the obtained three-dimensional aerodynamic stagnation interface is projected onto the cross-section of the main pipe to form a segmented curve. The area of ​​the weak-side fluid region enclosed by the projected curves of the main pipe's physical inner wall and the aerodynamic stagnation interface is calculated using numerical integration. Figure 3 The area marked as shaded in the diagram is considered the effective flow section. When calculating the wetted perimeter, it is combined with... Figure 3 The definition of the boundary is divided into two parts for calculation: one part is the arc length of the contact between the weak-side fluid and the physical pipe wall (corresponding to...). Figure 3 The other part is the boundary length of the interface between the weak-side fluid and the aerodynamic stagnation interface (corresponding to the physical inner wall of the mother tube). Figure 3 The values ​​of these two parts (the aerodynamic stagnation interface projection curve) are obtained by summing the curves of the physical pipe wall boundary segment and the aerodynamic interface projection curve segment on the calculated two-dimensional geometry of the cross-section. The effective hydraulic diameter of the weak-side fluid is calculated by dividing the effective flow cross-section by four times the wetted perimeter. The physical meaning of the effective flow cross-section and effective hydraulic diameter is: to quantify the degree to which the actual usable flow channel of the weak-side fluid is reduced due to the aerodynamic crowding effect of the strong-side fluid. In an asymmetric confluence flow field, the weak-side fluid does not fill the entire physical pipe, but is forced to flow in a compressed virtual pipe. The effective flow cross-section and effective hydraulic diameter accurately describe the geometry of this virtual pipe and are directly related to the friction loss and local pressure loss of the fluid.

[0089] By constructing an aerodynamic stagnation interface model and calculating effective hydraulic parameters, the fluid dynamics root cause of flow measurement distortion under asymmetric conditions can be revealed. Traditional methods often ignore the changes in the flow field structure caused by fluid collisions and assume that the fluid fills the pipe. However, this implementation method, by quantifying the spatial compression effect of the strong-side fluid on the weak-side fluid, proves that the weak-side fluid faces an actual flow capacity smaller than the physical pipe diameter. This geometric reconstruction method based on momentum balance simplifies the complex turbulent collision problem into a computable geometric parameter change problem, providing a correction factor with clear physical meaning for subsequent flow coefficient correction. Thus, without adding hardware sensors, the accuracy of steam metering in asymmetric confluence flow fields is improved at the algorithmic level.

[0090] Step S2222: Using the aerodynamic stagnation interface surface as the initial boundary condition for shear mixing, and combining the Kelvin-Helmholtz instability characteristics caused by the momentum flux ratio, construct a spatial distribution model of the shear mixing layer extending along the parent tube axis.

[0091] The physical spatial coordinates of the temperature sensor on the main pipe are projected into the spatial distribution model of the shear mixing layer, and the geometric deviation vector between the sensor probe position and the true energy centroid of the fluid is calculated. Combining the correction effect of the effective hydraulic diameter on flow measurement and the correction effect of the geometric deviation vector on enthalpy measurement, the physical feature set of flow field distortion is generated.

[0092] The process of constructing a spatial distribution model of the shear mixing layer extending along the axial direction of the parent pipe is as follows: Read the aerodynamic stagnation interface surface data determined in the previous steps. Specifically, the surface data is the set of spatial coordinates of all aerodynamic stagnation interface grid nodes obtained after the convergence of the previous iterative solution. Use the aerodynamic stagnation interface surface as the central skeleton surface of the shear mixing layer. This skeleton surface extends in the Z-axis direction. Based on the Kelvin-Helmholtz instability principle, since there is a velocity difference between the strong-side fluid and the weak-side fluid on both sides of the aerodynamic stagnation interface, this velocity shear will cause the interface to generate vortices and gradually break up and mix, forming a mixing region with a certain thickness. The theoretical basis of the calculation method stems from Prandtl's mixing length theory and the self-similarity characteristics of the turbulent free shear layer: According to the turbulent boundary layer theory, in a free shear layer formed by two parallel fluids, due to the entrainment effect of turbulent vortices, the lateral width of the mixing layer increases linearly with the flow direction distance. Simultaneously, according to the momentum transfer principle, the expansion rate of the mixing layer is directly related to the shear strength of the two fluids, which is physically determined by the ratio of the velocity difference between the two fluids to the average convection velocity. Therefore, based on dimensional analysis, the mixing layer thickness is mathematically derived as the product of the dimensionless structural expansion rate, the dimensionless kinematic velocity ratio, and the geometric flow direction length. Based on this theoretical support, the thickness of the shear mixing layer is calculated using the velocity difference between the strong-side and weak-side fluids on both sides of the aerodynamic stagnation interface. The calculation method for the shear mixing layer thickness is as follows: taking the aerodynamic stagnation interface as a reference, extending in the normal direction to both sides of the interface, the width of the extension is determined by the product of the expansion coefficient, the distance along the flow direction, and the velocity ratio factor. The distance along the flow direction refers to the axial length extending downstream from the initial confluence point of the aerodynamic stagnation interface upstream of the main pipe, along the axis of the main pipe, to the current calculated section. The experimentally determined expansion coefficient is obtained through pre-conducted standard fluid dynamics wind tunnel experiments or high-precision numerical simulations. Under similar Reynolds number and momentum-flux ratio conditions, the slope of the free shear layer width increasing with the flow direction distance is measured as this coefficient. The velocity ratio factor is the ratio of the absolute value of the difference between the strong-side fluid velocity and the weak-side fluid velocity to their sum. The spatial region enclosed by the extended upper and lower boundary surfaces constitutes the spatial distribution model of the shear mixing layer. The physical meaning of the shear mixing layer spatial distribution model is that it defines the transition region where substantial mass exchange and energy transfer occur between the two fluids. Within this transition region, the physical properties of the fluids, such as temperature and velocity, exhibit a gradient distribution characteristic from strong-side properties to weak-side properties, rather than the properties of a single fluid.

[0093] The steps for calculating the geometric deviation vector are as follows: First, obtain the physical installation coordinates of the temperature sensor mounted on the main pipe and project these coordinates onto the established spatial distribution model of the shear mixing layer. Then, calculate the true energy centroid of the fluid, defined as the weighted geometric center of enthalpy flux on the cross-section of the main pipe. During calculation, a weighted integral is performed on the fluid region on the cross-section where the sensor is located. Weighting factors include fluid density, fluid velocity, and specific enthalpy. The product of density, velocity, specific enthalpy, and the position vector is integrated on the cross-section and then divided by the integral value of the product of density, velocity, and specific enthalpy to obtain the spatial coordinates of the true energy centroid of the fluid. Specifically, the specific enthalpy is obtained by consulting the standard thermodynamic property table (such as the IAPWS-IF97 standard) published by the International Association for the Properties of Water and Steam (IAWAS) based on the local pressure field and temperature field data obtained from fluid dynamics simulation. The position vector refers to the spatial coordinate vector of the fluid particle relative to the geometric center of the cross-section of the main pipe. Because the strong-side fluid has a higher proportion of momentum and flow rate, the calculated true energy centroid is usually biased towards the strong-side fluid region. A vector pointing from the temperature sensor probe position to the true energy centroid of the fluid is constructed; this is the geometric deviation vector. The physical meaning of the geometric deviation vector is that it quantifies the sampling position error of a single-point temperature sensor in a non-uniform temperature field. The magnitude and direction of this vector directly reflect the degree to which the measurement point deviates from the average energy state of the fluid, revealing the spatial geometric reason why temperature measurements cannot represent the average temperature of the cross-section.

[0094] The steps for generating the physical feature set of flow field distortion are as follows: Extract the effective hydraulic diameter and momentum-flux ratio of the weak-side fluid, representing the degree of velocity distribution distortion, from the previous steps, and the geometric deviation vector, representing the degree of temperature distribution distortion, calculated in this step. To eliminate the influence of different physical dimensions, the momentum-flux ratio, effective hydraulic diameter, modulus of the geometric deviation vector, and direction angle of the geometric deviation vector are normalized. The normalization process uses the range transformation method, mapping the values ​​of each parameter to a dimensionless range of zero to one. Specifically, the current parameter value is subtracted from the minimum value of that parameter in the historical dataset, and then divided by the difference between the maximum and minimum values. The normalized parameters are combined to form the physical feature set of flow field distortion. The physical significance of this feature set is that it decouples the hydrodynamic factors causing measurement errors in complex asymmetric confluence flow fields into two independent geometrical-physical features: "narrowing of the flow cross-section" and "shifting of the energy center," providing standardized input variables for the subsequent construction of a high-dimensional flow coefficient correction model.

[0095] By establishing a shear mixing layer model and calculating the geometric deviation vector, a quantitative description of the non-uniform energy distribution in an asymmetric flow field was achieved. The Kelvin-Helmholtz instability characteristics were used to determine the mixing layer range, ensuring that the calculation of the transition region of fluid properties conforms to the physical laws of turbulent mixing. By calculating the deviation between the sensor position and the true energy centroid, the sampling representativeness error caused by fluid stratification was accurately captured. The final generated set of physical features of flow field distortion transforms the flow field structural distortion at the fluid dynamics level into geometric feature parameters that can participate in numerical calculations. This allows for the elimination of metering deviations caused by fluid collisions through algorithmic compensation without altering the existing pipeline structure and sensor layout, thus improving the accuracy and reliability of steam metering in complex pipe networks.

[0096] Step S223: Using the physical feature set of flow field distortion as the independent variable and the baseline prediction residual as the dependent variable, construct an asymmetric load correction factor regression model, analyze the contribution weight of physical distortion to measurement deviation, and output an asymmetric load correction factor that includes fluid dynamic constraints.

[0097] The input layer vector and output layer objective of the asymmetric load correction factor regression model are constructed as follows: Four key parameters are extracted from the flow field distortion physical feature set generated in the previous step as input variables of the model. These four parameters are: normalized momentum-flux ratio, normalized effective hydraulic diameter of the weak-side fluid, normalized geometric deviation vector magnitude, and normalized geometric deviation vector direction angle. The momentum-flux ratio is obtained by calculating the ratio of the momentum flux of the strong-side fluid to that of the weak-side fluid, reflecting the difference in impact penetration ability of the two fluids at their convergence. The other three parameters are obtained in the same way as in the previous step. All input parameters have been normalized using the range transformation method to eliminate dimensional differences. The training objective of the model is set as the baseline prediction residual, which is the relative error between the flow rate value calculated based on the uncorrected flow coefficient in the previous step and the actual flow rate value measured by a high-precision standard meter. When constructing the dataset, historical experimental data or high-fidelity simulation data covering different flow rates, temperature stratifications, and valve opening combinations are selected as training samples to form a mapping pair between the input feature vector and the output residual target. The physical meaning of this step is: to associate the discretized measurement errors determined in step S221 with the specific fluid dynamic causes that lead to these errors through mathematical modeling, thereby enabling the system to infer the measurement errors based on physical characteristics under unknown operating conditions.

[0098] The specific architecture of the asymmetric load correction factor regression model employs a backpropagation neural network with a single or multiple hidden layers, which serves as a nonlinear regressor. The input layer has four nodes, corresponding to the four normalized physical features mentioned above; the output layer has one node, corresponding to the baseline prediction residual prediction value. The number of neurons in the hidden layers is determined using empirical formulas combined with trial and error. Hyperbolic tangent or linear rectified functions are chosen as activation functions to introduce nonlinear mapping capabilities. The model training process uses the Levenberg-Marquardt algorithm or gradient descent, iteratively adjusting the inter-layer connection weights and thresholds to minimize the mean square error between the predicted residual and the true residual. After training, the contribution weights of each physical feature to the measurement bias are analyzed using the connection weight method. Specifically, the absolute values ​​of the products of the connection weights from the input layer to the hidden layer and the connection weights from the hidden layer to the output layer are summed. This sum is used as an indicator of the importance of the corresponding input feature; a larger value indicates a more significant impact of the physical distortion feature on the measurement bias. The technical objective of this process is to utilize the powerful nonlinear approximation capability of neural networks to fit a highly nonlinear functional relationship between flow field distortion characteristics and measurement errors, and to quantify and separate the specific contributions of different fluid dynamic factors to the error generation from the black box model.

[0099] The process of outputting the asymmetric load correction factor containing fluid dynamic constraints is as follows: The physical feature set of flow field distortion obtained from real-time acquisition and calculation is input into the trained nonlinear compensation model, and the model outputs the estimated residual value under the current operating condition. The asymmetric load correction factor is calculated based on the error inversion principle. Since the estimated residual value is defined as the relative deviation of the measured value from the true value, in order to restore the true value, the original flow coefficient needs to be inversely compensated. Therefore, the calculation formula for the asymmetric load correction factor is set as the sum of the standard unit value 1 divided by 1 and the estimated residual value. When the measured value is greater than the true value, the residual is positive; otherwise, it is negative. This correction factor is used as the final output and is directly multiplied by the original flow coefficient in the flow calculation stage. The physical meaning of this step is: to convert the additive error predicted by the model into a multiplicative correction coefficient, and to offset the systematic positive and negative deviations caused by flow field distortion through mathematical division, ensuring that the corrected flow coefficient can truly reflect the average flow velocity characteristics of the fluid.

[0100] By constructing a nonlinear compensation model with normalized flow field characteristics as input, high-precision prediction and correction of metering deviations in complex converging flow fields were achieved. A multi-dimensional feature set, including momentum flux ratio and geometric deviation vector, was employed to comprehensively cover various error sources from fluid impact dynamics to sensor spatial sampling geometry, ensuring the model's generalization ability under varying operating conditions. By analyzing the contribution weights of each feature, the dominant factors causing errors could be identified, providing data support for subsequent optimization of pipeline design or sensor layout. The final output asymmetric load correction factor can dynamically adapt to transient changes in the flow field structure in real time. Without changing the hardware, the algorithm-level soft measurement technology significantly improves the metering accuracy of the heating network system under off-design conditions, solving the technical challenge of traditional single-point measurement failure in non-uniform flow fields.

[0101] Step S23: Apply the asymmetric load correction factor to the actual measured data of steam metering and thermodynamic state corresponding to the asymmetric operating section, use a robust regression iterative reweighting strategy to suppress outliers, and output the equivalent steam flow set and the equivalent thermodynamic state set.

[0102] The specific referent of the actual measurement data for the asymmetric section is clarified. This data directly originates from the steam metering thermodynamic state vector defined in step S21, specifically including: the main steam flow rate of the first waste heat boiler and the main steam flow rate of the second waste heat boiler, collectively referred to as the flow rate sequence; and the main steam temperature of the first waste heat boiler, the main steam temperature of the second waste heat boiler, and the combined pressure of the shared header, collectively referred to as the thermodynamic state sequence. The asymmetric load correction factor output in step S223 is read and denoted as... This factor is essentially a correction coefficient for the momentum transport characteristics of fluids.

[0103] For flow sequences, a full multiplicative correction strategy is employed. This is because the measurement principle of flow meters (such as differential pressure or vortex flow meters) is highly dependent on the distribution of the velocity profile. Flow field distortion directly leads to a systematic deviation in the measured values ​​that is proportional to the degree of distortion. Therefore, each instantaneous sample value in the flow sequence is directly multiplied by an asymmetric load correction factor. This is done to completely offset the measurement distortion caused by changes in the effective flow cross section and momentum compression, resulting in a preliminarily corrected flow sequence.

[0104] For the thermodynamic state sequence (temperature and pressure), a weighted incremental compensation strategy based on distortion sensitivity is adopted. The physical reason for adopting this strategy is that temperature and pressure belong to the fluid scalar field, while flow rate belongs to the vector field. Although fluid collisions and shearing effects (i.e., flow field distortion) within the main pipe can drastically change the velocity distribution, their impact on pressure conduction and temperature diffusion is relatively small. The influence of flow field distortion on temperature measurement mainly stems from the geometric deviation vector calculated in step S2222, i.e., the sensor probe position deviates from the true energy center of the fluid. If a correction factor for flow rate is used directly... Applying full corrections to temperature and pressure can lead to overcompensation, causing thermodynamic parameters to deviate significantly from their physical true values. For example, a 10% correction for flow rate is reasonable, but a 10% correction for temperature violates thermodynamic principles. The distortion sensitivity coefficient is obtained and calculated by introducing a distortion sensitivity coefficient... The magnitude of the geometric deviation vector is positively correlated with the normalized geometric deviation vector generated in step S2222. That is, the larger the magnitude of the geometric deviation vector, the further the sensor is from the energy center, and the greater the influence of flow field distortion on temperature / pressure. The larger the value, the more likely it is to be true. The value ranges from 0.05 to 0.2. The calculation formula is: In the formula, The original temperature or pressure measurement value. This is the value after preliminary correction. The physical meaning of this formula is: extract the distortion portion that deviates from the unit 1 in the correction factor. Using sensitivity coefficient By attenuating the flow field distortion, only the limited influence of the flow field distortion on the scalar field is superimposed back onto the original measurement value, thus obtaining a preliminary corrected thermodynamic state sequence.

[0105] The preliminarily corrected flow rate and thermodynamic state sequences were processed using an iterative reweighting strategy based on robust regression to suppress outliers, eliminating non-physical outliers caused by fluid turbulence fluctuations or electromagnetic interference. The implementation of the iterative reweighting strategy based on robust regression for outlier suppression is as follows: The preliminarily corrected flow rate sequences are cleaned to eliminate sensor random noise and non-physical jumps caused by transient fluctuations in the flow field. The median of the sequence is selected as the initial estimate, and the residual between each data point and the initial estimate is calculated. The weight of each data point is calculated using the residuals, employing a double-squared weighting function: an optimization constant is set. The standard value is typically 4.685 times the median absolute deviation (MAD) of the residuals. For residuals with an absolute value less than... The weight calculation formula for the data points is as follows: In the formula, The standardized residual represents the distance between the current data point and the estimated value. This is used to fine-tune the threshold constant, defining the boundary between normal fluctuations and outliers. For residuals with an absolute value greater than or equal to... Data points are considered outliers, with their weights reset to 0. Based on the calculated weights, the weighted mean of the sequence is recalculated using weighted least squares as a new estimate. This process of calculating residuals, updating weights, and recalculating the weighted mean is repeated until the difference between the weighted means calculated in two consecutive iterations is less than a preset convergence threshold, for example... The weighted mean sequence at this point is the equivalent data after smoothing. The physical significance of this step lies in introducing residuals... and tuning constants The comparison mechanism allows the algorithm to automatically identify the degree of data dispersion. When the fluid undergoes normal operating condition adjustments, the residuals are within a reasonable range, and the data is retained; however, when instantaneous measurement jumps occur due to electrical spikes or localized fluid cavitation, the residuals... It will increase significantly and exceed This results in zero weighting, enabling precise removal of non-physical noise. Statistical methods are used to identify and eliminate abrupt data that does not conform to the fluid continuity characteristics, ensuring that the output values ​​reflect the steady-state or quasi-steady-state characteristics of the fluid, rather than spurious signals caused by transient turbulence or electrical interference. This guarantees that subsequent energy efficiency calculations are based on physically real fluid energy states, rather than sensor measurement artifacts. The processed flow sequence is output as an equivalent steam flow set, and the processed temperature and pressure sequence is output as an equivalent thermodynamic state set. These two sets reconstruct the true thermodynamic parameters that the unit should exhibit under asymmetric heating conditions, assuming no flow field distortion and uniform mixing. This provides clean and reliable data input for constructing virtual equivalent energy channels and calculating the overall plant thermal efficiency in subsequent step S30.

[0106] Step S30: Using the equivalent steam flow set and the equivalent thermodynamic state set as input, construct a virtual equivalent energy channel for the shared main pipe and heating extraction steam topology, calculate the overall plant thermal efficiency, generate energy efficiency reliability characterization quantity, and output a reliable energy efficiency index set.

[0107] Step S30 includes the following sub-steps:

[0108] Step S31: Using the equivalent steam flow set and the equivalent thermodynamic state set as input, generate a virtual main pipe thermodynamic state based on the shared main pipe topology relationship, so that the contribution of the two waste heat boiler sides to the turbine steam inlet and heating extraction steam can be allocated in the digital domain.

[0109] Based on the thermal state of the virtual main pipe, the equivalent energy quantification results of the power generation output channel, the heat supply output channel and the unavailable loss channel are calculated to form an equivalent energy channel set.

[0110] The implementation method for generating the virtual header thermal state is as follows: retrieve the equivalent steam flow set and equivalent thermal state set output in step S23, wherein the equivalent steam flow set includes the corrected main steam flow of the first waste heat boiler. Main steam flow rate of the second waste heat boiler The equivalent thermodynamic state set includes the corrected main steam temperature of the first waste heat boiler. The main steam temperature of the second waste heat boiler and the shared main pipe combined pressure Based on the physical topology of the shared main pipe, i.e., the Y-shaped node where two branch pipes converge into one main pipe, an energy conservation balance equation is established. The IAPWS-IF97 international standard formulas for the properties of water and water vapor are then applied. and Calculate the specific enthalpy of the first steam stream. ,use and Calculate the specific enthalpy of the second steam stream. Assuming the two fluids undergo adiabatic mixing within the header pipe, calculate the enthalpy of the virtual header pipe mixing. The calculation formula is: Based on this mixing enthalpy, an energy contribution distribution coefficient is defined, specifically the distribution coefficient on the first waste heat boiler side. for Distribution coefficient on the second waste heat boiler side for The physical significance of generating the virtual master tube thermal state is as follows: it reconstructs the homogenized energy state after fluid mixing in the digital space, and mathematically decouples and restores the total energy flow that has been irreversibly mixed by fluid turbulence to the independent contribution ratio of each heat source side through enthalpy weighting, thereby establishing the calculation benchmark for energy traceability of multi-axis units.

[0111] The implementation method for calculating the equivalent energy quantization result of the power generation output channel is as follows: obtain the real-time total active power of the turbine generator set from the distributed control system. Using the allocation coefficients obtained from the aforementioned calculations, the total active power is decomposed, and the equivalent power generation corresponding to the first gas turbine side is calculated. and the equivalent power generation corresponding to the second gas turbine side. .Will and As a quantitative result of the power generation output channel, the physical significance of this step is: to establish a mapping channel from the heat energy input end to the electrical energy output end, and to accurately assign the total electrical energy generated by the steam turbine, a shared power-working component, to its respective combustion drive source according to the energy conservation principle of the first law of thermodynamics and the weighted contribution ratio of the enthalpy value of the steam input from each heat source side, thereby realizing the independent evaluation of the power generation performance in the two-on-one unit.

[0112] The method for calculating the equivalent energy quantification result of the heating output channel is as follows: obtain the steam extraction flow rate on the heating extraction pipeline. extraction steam temperature and extraction pressure And the return water temperature on the heating network return water pipe and return water pressure Calculate the extraction enthalpy using the IAPWS-IF97 standard. Enthalpy of return water Calculate the total heat energy supplied by the entire plant to external parties. Based on the enthalpy-weighted contribution ratio of the input energy from each heat source side, the total heat energy is distributed to each side using a distribution coefficient, i.e., the equivalent heat supply to the first gas turbine side. Equivalent heat supply from the second gas turbine side .Will and As a quantitative result of the heat output channel, it can quantify the actual load-bearing capacity of each unit on the external heating network under combined heat and power (CHP) conditions, eliminating the ambiguity in heat metering attribution caused by steam extraction location and mixing effects.

[0113] The implementation method for calculating the equivalent energy quantification results of unavailable loss channels and forming a set is as follows: A plant-wide energy balance model is established based on the first law of thermodynamics. The total steam energy of the input system is calculated. Calculate the total energy loss of the entire plant. This loss includes cold source loss, mechanical friction loss, and heat dissipation loss. Similarly, the equivalent loss shared by each side is calculated using a distribution factor. , The equivalent power generation, equivalent heat supply, and equivalent losses calculated above for each side are categorized according to the corresponding units and combined to form an equivalent energy channel set. This closes the energy balance chain, makes the system dissipation energy, which cannot be directly measured, explicit and reasonably allocated, providing complete energy flow data support for subsequent calculations of the plant's overall thermal efficiency.

[0114] By constructing a virtual header and equivalent energy channels, this implementation method successfully solves the energy decoupling problem in multi-heat-source coupled systems without adding physical isolation equipment. Using high-confidence data corrected for flow field distortion as input, combined with enthalpy-weighted allocation logic, it ensures that the allocation results conform to the physical reality of energy grade differences. This digital allocation method enables operators to clearly identify the actual energy efficiency contribution of a single gas turbine in combined cycle mode, providing a quantitative basis for load optimization allocation under asymmetric operating conditions and improving the level of refined unit management.

[0115] Step S32: Calculate the overall plant thermal efficiency from the equivalent energy channel set, and generate the energy efficiency reliability characterization quantity based on the reconstructed confidence interval, cross-channel consistency verification results and segment stability, and output the reliable energy efficiency index set.

[0116] The method for calculating the overall plant thermal efficiency is as follows: Extract the equivalent power generation and equivalent heat supply values ​​for the first and second gas turbine sides from the equivalent energy channel set generated in step S31; synchronously obtain the natural gas fuel consumption and lower heating value of the corresponding gas turbine from the distributed control system. Based on the first law of thermodynamics, add the equivalent power generation and equivalent heat supply values ​​to obtain the total effective energy output; multiply the natural gas fuel consumption and the lower heating value of the natural gas to obtain the total fuel energy input; divide the total effective energy output by the total fuel energy input and convert it to a percentage to obtain the overall plant thermal efficiency for each side. The energy efficiency index calculated here is a net value after eliminating the influence of flow field distortion and completing energy allocation. Its physical meaning is: it truly reflects the actual heat-work conversion capacity of a single gas turbine and its corresponding waste heat utilization system under the current asymmetric operating conditions, eliminating false energy efficiency fluctuations caused by metering errors.

[0117] Generate energy efficiency reliability metrics: Construct a comprehensive evaluation model encompassing three dimensions: reconstructed confidence interval, cross-channel consistency verification results, and segment stability. The following evaluation process is executed independently for the first and second gas turbine sides. Calculate the reconstructed confidence interval index: Retrieve the prediction variance output from the nonlinear compensation model in step S23 and the residual scaling estimate from the robust regression process. Using the t-distribution theory in statistics, calculate the confidence interval width of the corrected equivalent steam flow rate at a 95% confidence level. Divide this confidence interval width by the current equivalent steam flow rate value to obtain the relative uncertainty. A reverse mapping function is constructed. When the relative uncertainty approaches zero, the reconstructed confidence interval index approaches one; when the relative uncertainty exceeds a preset engineering tolerance limit, the reconstructed confidence interval index approaches zero. The engineering tolerance limit is usually set to five percent. In this embodiment, the reverse mapping function specifically adopts a negative exponential decay function model. The operation logic of the negative exponential decay function model is as follows: using the natural constant as the base and the negative value of the product of the relative uncertainty and the preset sensitivity decay coefficient as the exponent, the reconstructed confidence interval index is calculated. The sensitivity attenuation coefficient is determined by back-calculation using boundary conditions. Specifically, when the relative uncertainty reaches the engineering allowable limit, the calculated index value should attenuate to a preset low confidence cutoff threshold. The low confidence cutoff threshold is determined as follows: The hardware specifications of the flow sensor configured in the distributed control system are read, the full-scale accuracy percentage parameter is extracted, and it is multiplied by the current measurement range. The resulting product is determined as the maximum permissible measurement error. The historical operating sequence at the end of the previous maintenance cycle (i.e., when equipment performance degrades to a critical maintenance state) is extracted and input into the benchmark mapping model. The deviation set between the theoretical predicted value sequence and the actual measured value sequence is calculated, and this deviation set is determined as the benchmark residual distribution. The expected value and variance of the baseline residual distribution are calculated. Using the normal distribution probability density function, the probability density integral value is calculated when the deviation value equals the maximum permissible measurement error, indicating that the value falls within the baseline residual distribution interval. This probability density integral value is extracted and determined as the low confidence cutoff threshold. The technical advantage of the inverse mapping function form is that when the relative uncertainty is small, the index value remains relatively stable at a high level close to one, reflecting tolerance for small calculation errors. However, once the relative uncertainty increases, the index value will drop sharply, reflecting a strict penalty for data quality deterioration. Furthermore, this invention is not limited to this; any function model that satisfies monotonically decreasing and takes a value of one at zero can be used as an alternative. The physical meaning of the reconstructed confidence interval index is: it quantifies the reliability of the data correction algorithm at the mathematical and statistical level, characterizing the probability that the corrected traffic data falls near the true physical value. The cross-channel consistency verification result index is calculated by retrieving the equivalent loss value of the unavailable loss channel calculated in step S31.Consult the unit design specifications or heat balance diagram to obtain the theoretical design loss value at the current load rate. Calculate the absolute value of the difference between the equivalent loss value and the theoretical design loss value, and then divide it by the theoretical design loss value to obtain the relative deviation. Use an exponential decay function to convert the relative deviation into a consistency score, i.e., a cross-channel consistency verification result index; this exponential decay function includes a consistency penalty sensitivity coefficient, the specific determination rule of which is as follows: extract continuous operating data of the unit under pure power generation and no heating steam extraction operation conditions; based on the aforementioned plant-wide energy balance model based on the first law of thermodynamics, calculate the difference between the total steam energy of the input system and the total active power of the turbine generator unit in each sampling period within this time period; sort the differences by time to obtain the unavailable loss value sequence; calculate the statistical variance of this unavailable loss value sequence to determine the background fluctuation variance; obtain the ASME PTC... The maximum permissible uncertainty constant for system-level heat rate testing, as defined in the "46 Combined Cycle Unit Performance Test Procedure," is determined as the maximum permissible relative error for system heat rate measurement. The square of this maximum heat rate measurement relative error is calculated, and this square is divided by the background fluctuation variance. The resulting quotient is determined as the consistency penalty sensitivity coefficient. This rule ensures that when the deviation in energy balance calculation crosses the background noise band, the algorithm uses this coefficient to drive an exponential decay in the consistency score, controlling the rate at which the score decreases with increasing deviation. The technical purpose of the cross-channel consistency verification result index is to use the law of conservation of energy as a physical constraint to verify whether the corrected data conforms to the macroscopic energy balance characteristics of the thermal system, preventing calculation results that, while mathematically well-fitted, violate physical principles. The stability index for the calculation section is as follows: Within the current calculation time window, the plant-wide thermal efficiency sequence is extracted, and the standard deviation and arithmetic mean of the sequence are calculated. The coefficient of variation is obtained by dividing the standard deviation by the arithmetic mean. A stability threshold is set, which is taken from the energy efficiency fluctuation rate measured during the unit's historical steady-state performance tests. The ratio of the coefficient of variation to the stability threshold is calculated by subtracting the numerical value from the calculated value. If the result is less than zero, it is set to zero. This result is the section stability index. The section stability index evaluates the stability of the current operating condition in the time domain, identifying and reducing the interference of transient processes or violent oscillations on the weight of energy efficiency assessment conclusions. A comprehensive energy efficiency reliability characterization quantity is generated based on the above three sub-indicators. A weighted summation method is used, multiplying the reconstructed confidence interval index, the cross-channel consistency verification result index, and the section stability index by their respective weighting coefficients and then summing them. The weighting coefficients are determined based on expert experience or the analytic hierarchy process, and the sum of the three weighting coefficients is one. The calculated plant-wide thermal efficiency value is combined with the corresponding energy efficiency reliability characterization quantity to form a reliable energy efficiency index set.

[0118] By introducing a multi-dimensional credibility evaluation mechanism, the single numerical energy efficiency index is upgraded to a composite index that includes both numerical and qualitative dimensions. This approach not only provides high-precision energy efficiency data corrected for flow field distortion but also simultaneously provides self-checking results for the data's reliability. This allows operators or upper-level optimization systems to determine the degree of data acceptance based on its credibility. Under complex operating conditions such as asymmetric operation, this index set can effectively filter out spurious data generated by model extrapolation or system disturbances, ensuring the rigor and anti-interference capability of the unit performance evaluation system.

[0119] Step S40: Write the set of reliable energy efficiency indicators into the indicator point table of the distributed control system, and generate pseudo-high energy efficiency alarm and fluctuation suppression display parameters based on the energy efficiency reliability characterization quantity.

[0120] Step S40 includes the following sub-steps:

[0121] Step S41: Map the set of reliable energy efficiency indicators to indicator points that the distributed control system can subscribe to, and output the quality label corresponding to the calculation cycle;

[0122] Based on the energy efficiency credibility representation quantity and the preset pseudo-high energy efficiency rule base, it is determined whether the current energy efficiency display is falsely high;

[0123] The reliable energy efficiency index set is mapped to index points that the distributed control system can subscribe to: an OPC UA or Modbus TCP communication connection is established between the energy efficiency analysis server and the power plant's distributed control system. Independent storage address spaces are pre-allocated in the real-time database of the distributed control system, corresponding to the calculation results from the first gas turbine side and the second gas turbine side, respectively. Specific mapping objects include: the overall plant thermal efficiency of the first gas turbine side, the energy efficiency reliability characterization quantity of the first gas turbine side, the overall plant thermal efficiency of the second gas turbine side, the energy efficiency reliability characterization quantity of the second gas turbine side, and the timestamp of data generation. The values ​​in the above index set are written to the corresponding storage addresses in real time. The implementation method for the quality flag corresponding to the calculation cycle is as follows: within each calculation cycle, the integrity and timeliness of the original collected data are monitored. If the sampling frequency of all sensor data used for calculation in the current cycle satisfies the Nyquist sampling theorem, and the packet loss rate during data transmission is less than 0.1%, then the quality flag is set to a Boolean value of true, indicating that the data is valid; otherwise, it is set to a Boolean value of false, indicating that the data is invalid. The technical objective of this step is to establish an information exchange channel between the energy efficiency calculation edge and the power plant's main control system, ensuring that operators can obtain high-precision energy efficiency data after flow field distortion correction in a unified monitoring screen, and to use quality markers to prevent outdated data from misleading decision-making due to communication failures or missing data.

[0124] The implementation method for constructing a pre-defined pseudo-high-efficiency rule base is as follows: The rule base includes thermodynamic limit rules and low-load high-efficiency anomaly rules. The thermodynamic limit rules are set based on: querying the heat balance design specifications of the gas turbine and combined cycle unit to obtain the maximum design thermal efficiency value of the unit under ISO standard operating conditions; considering measurement uncertainty and the slight performance improvement that equipment modifications may bring, this maximum design thermal efficiency value is multiplied by 1.05 to obtain the physical limit threshold. The low-load high-efficiency anomaly rules are set based on: using the unit's historical operating data from the past three years, a polynomial fitting method is used to construct a benchmark characteristic curve of load rate and thermal efficiency, determining the upper bound of the expected efficiency range below 50% load rate; this upper bound of the expected efficiency range is multiplied by 1.1 to obtain the low-load anomaly threshold. The physical significance of this rule base construction is: using the objective energy conversion limit determined by the second law of thermodynamics, and the inherent variable operating condition characteristic of the gas turbine's efficiency inevitably decreasing under partial load, a reasonable boundary that energy efficiency values ​​may physically exist is established, providing a deterministic physical criterion for identifying false data.

[0125] The implementation method for determining whether the current energy efficiency display is falsely inflated based on the energy efficiency reliability characterization quantity and the preset pseudo-high energy efficiency rule base is as follows: Independent discrimination logic is executed for the first gas turbine side and the second gas turbine side respectively. The energy efficiency reliability characterization quantity generated in step S32 is read, and the acceptance threshold is set to 0.6, which is determined according to the definition of the low confidence interval in statistics. The discrimination logic is as follows: First, the currently calculated plant-wide thermal efficiency value is compared with the physical limit threshold in the thermodynamic limit rule. If it exceeds the threshold, it is determined that there is a false inflated value. Second, the current unit load rate is checked. If the load rate is less than 50% and the plant-wide thermal efficiency value exceeds the low load anomaly threshold, it is determined that there is a false inflated value. Finally, the energy efficiency reliability characterization quantity is checked. If the value is lower than the acceptance threshold, it indicates that the calculation result is greatly affected by the extrapolation of the flow field distortion correction model or sensor noise, and it is also determined that there is a false inflated value. As long as any of the above conditions are met, it is determined that the current energy efficiency display is falsely inflated. The technical objective of this step is to construct a dual logic filter by combining physical rule constraints with statistical confidence assessment. This filter identifies calculation results that, while valid in mathematical regression calculations, violate physical common sense or have low statistical confidence, thus preventing the illusion of high energy efficiency caused by sensor drift, extreme flow field distortion, or model overfitting.

[0126] By constructing a deep mapping and interaction strategy between a set of reliable energy efficiency indicators and a distributed control system, the system achieves refined management and closed-loop application of performance data from both units. When there is a single-side instrument failure, severe single-side flow field distortion, or a single-side unsteady-state transition, the system can identify the inferior data on that side based on quality markers and pseudo-high energy efficiency identification results, ensuring the authenticity of the data. The pseudo-high energy efficiency identification function based on physical rules effectively prevents false data from entering the operation monitoring stage, ensuring the rigor of the unit performance evaluation system and providing operators with high-quality decision-making basis that has undergone rigorous physical and statistical verification.

[0127] Step S42: When the system is identified as having pseudo-high energy efficiency, an alarm signal and visual suppression parameters are generated, and operational recommendations for asymmetric heating regulation are output for operators to refer to.

[0128] Generating alarm signals and visual suppression parameters: When the false high energy efficiency judgment result output in step S41 is true, the alarm logic of the distributed control system is immediately triggered. The specific operation of generating the alarm signal is to write a high-priority alarm record to the alarm event list of the distributed control system, including the unit number where the false high efficiency occurred, the millisecond-level timestamp of the trigger time, and the trigger rule type. The operation of generating visual suppression parameters includes redefining color attributes and switching the value source: the background color attribute of the corresponding energy efficiency value displayed in the human-machine interface is changed from green (RGB value 0,255,0) representing normal efficiency to gray (RGB value 128,128,128) representing invalid efficiency, while the transparency parameter of the foreground value is adjusted to 50%; the data source for the value display is switched from the real-time calculated plant-wide thermal efficiency to the physical limit threshold determined in step S41, and a downward-pointing trend arrow is added next to the value. The technical objective of this step is to: through forced intervention via the visual channel, block the psychological suggestion that erroneous high-efficiency data may have on operators; reduce the salience of the data by using grayscale and transparency; and intuitively inform operators of the current physical boundaries by displaying physical limit thresholds, thereby preventing production accidents caused by blindly pursuing falsely high indicators.

[0129] The implementation method for outputting operational recommendations for asymmetric heating regulation is as follows: Based on the operating characteristics of the dual-unit combined cycle, when a false high-efficiency display is detected on one side of the gas turbine, an asymmetric load balance analysis is initiated. Real-time active power data of the first and second gas turbines are collected, and the absolute value of the power deviation between the two sides is calculated. If the absolute value of the power deviation exceeds 5% of the rated load, and the side exhibiting the false high-efficiency display is the high-load side, then the flow field distortion is determined to be caused by uneven exhaust velocity distribution due to excessive asymmetric heating. At this time, the correction target value for the inlet guide vanes is calculated. The correction algorithm is as follows: using the current inlet guide vane opening as a baseline, subtract the adjustment step size. The adjustment step size is set based on: querying the gas turbine anti-surge control curve to obtain the minimum allowable guide vane backoff margin under the current operating conditions, and taking 10% of this margin as the adjustment step size, typically set to 0.5 degrees. The calculated correction target value is then pushed to the operator station as operational recommendations. Pseudo-high efficiency is often accompanied by severe turbulence or deflection in the exhaust flow field. By finely adjusting the opening of the inlet guide vanes on the high-load side, the angle of attack and velocity distribution of the turbine exhaust can be changed while ensuring the load requirements. The aerodynamic reconstruction effect can be used to alleviate the flow field distortion, thereby restoring the measurement sensor to the normal linear operating range.

[0130] When generating recommended operating parameters, a redistribution recommendation for the heating extraction steam flow rate also needs to be output. If the unit is in heating operation, the current total heating flow rate demand is calculated; based on the evaporation capacity of the waste heat boilers on both sides, the flow rate allocation weight is calculated according to the proportion of the energy efficiency reliability indicator. Specifically, the unit on the side without false high energy efficiency display (i.e., high data reliability) is set as the baseline side, and its heating extraction steam flow rate setting value is increased; the side with false high energy efficiency display is set as the subordinate side, and its heating extraction steam flow rate setting value is reduced accordingly, with the increase or decrease controlled within three percent of the current total flow rate. This three percent threshold is derived from hydraulic calculations based on the allowable range of heating network pressure fluctuations (usually ±0.02 MPa). The technical purpose of this step is to indirectly adjust the exhaust back pressure of the gas turbine by changing the extraction steam boundary conditions on the low-pressure side, using pressure difference changes to smooth out the local vortices in the flow field that cause falsely high energy efficiency, and eliminating the physical causes of measurement distortion from the perspective of thermodynamic system coupling.

[0131] By generating alarm suppression and proactive adjustment suggestions, a closed-loop control auxiliary system, from passive monitoring to proactive intervention, is constructed. The system not only identifies and filters false information at the data level, eliminating the risk of misleading information, but also delves deeper into the thermal-hydraulic root causes of measurement distortion, providing specific equipment adjustment parameters. This approach enables operators to maintain stable total unit output while dynamically correcting flow field distortions by implementing asymmetric adjustment suggestions, thus restoring energy efficiency monitoring data to true values ​​and ensuring the observability and controllability of the unit under complex and variable operating conditions.

[0132] It should be understood that although the steps in the flowcharts of the various embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the various embodiments may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least a portion of the sub-steps or stages of other steps.

[0133] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0134] The foregoing description is illustrative of the invention and should not be construed as limiting it. Although several exemplary embodiments of the invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the invention. Therefore, all such modifications are intended to be included within the scope of the invention as defined in the claims. It should be understood that the foregoing description is illustrative of the invention and should not be construed as limiting it to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.

Claims

1. A data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units, characterized in that, The method includes the following steps: The continuous time-series operation data of the gas-fired steam combined cycle unit is parsed into a sequence of thermodynamic vector line segments. The sequence of thermodynamic vector line segments is then sieved based on the physical limit of thermal inertia to output symmetrical and asymmetrical operation segments. Based on symmetrical and asymmetrical operating sections, the regression model of asymmetrical load correction factor is trained using historical operating samples and called to identify and suppress the distortion caused by abnormal steam metering and nonlinear mixing under asymmetrical heating conditions, and output equivalent steam flow set and equivalent thermodynamic state set. Using the equivalent steam flow set and the equivalent thermodynamic state set as input, a virtual equivalent energy channel oriented towards the shared main pipe and heating extraction steam topology is constructed, the overall plant thermal efficiency is calculated, an energy efficiency reliability characterization quantity is generated, and a reliable energy efficiency index set is output. The set of reliable energy efficiency indicators is written into the indicator point table of the distributed control system, and pseudo-high energy efficiency alarm and fluctuation suppression display parameters are generated based on the energy efficiency reliability characterization quantity.

2. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 1, characterized in that, The steps for outputting symmetrical and asymmetrical operating segments include: Continuous time-series operation data of the gas-fired steam combined cycle unit is acquired. The second derivative of the continuous time-series operation data is scanned to identify the physical inflection point of the load change rate. The physical inflection point is used as the feature anchor point, and the operation trajectory between two adjacent feature anchor points is discretized and decomposed to output a sequence of thermodynamic vector line segments containing heat absorption vector, steady-state vector and heat release vector.

3. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 2, characterized in that, The steps of outputting symmetrical and asymmetrical operating segments further include: Based on the thermal inertia physical limit of the unit's metal components, the minimum physical transition time of each segment in the thermal vector line segment sequence is calculated and compared with the geometric time to eliminate thermal transient segments. Based on the compared heat absorption vector, steady-state vector, and heat release vector, the time-domain closure and dual-machine coupling are determined. The sequence with balanced dual-machine load and forming a closed thermodynamic cycle is marked as a symmetrical operating segment, while the sequence without forming a closed cycle, with residual thermal stress, or with unbalanced dual-machine load is marked as an asymmetrical operating segment.

4. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 1, characterized in that, The steps of outputting the equivalent steam flow set and the equivalent thermodynamic state set include: A symmetrical operating section sample was selected as the benchmark sample to train a benchmark mapping model from the operating condition descriptor to steam metering and thermodynamic state, so as to obtain the statistical correlation before the failure of the linear superposition assumption of the shared main pipe. Based on the benchmark mapping model, the benchmark prediction residual under the asymmetric operating section is calculated, and the physical characteristics of the flow field distortion at the confluence of the main pipes are coupled to construct a nonlinear compensation mechanism driven by physical causes, and generate an asymmetric load correction factor.

5. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 4, characterized in that, The step of generating the asymmetric load correction factor further includes: Input the working condition descriptor of the asymmetric section into the benchmark mapping model, obtain the theoretical prediction value based on the symmetry assumption, and calculate the benchmark prediction residual between the theoretical prediction value and the actual effective measurement set. Based on the spatial geometric topology of the junction of the outlets of the two waste heat boilers and the main pipe, and combined with the real-time flow rate and density of the two steam streams, the momentum flux ratio and shear layer distribution during fluid collisions are deduced, and a set of physical characteristics of flow field distortion is generated. Using the physical feature set of flow field distortion as the independent variable and the baseline prediction residual as the dependent variable, an asymmetric load correction factor regression model is constructed to analyze the contribution weight of physical distortion to measurement deviation and output an asymmetric load correction factor that includes fluid dynamic constraints.

6. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 5, characterized in that, The steps for generating the physical feature set of flow field distortion include: A three-dimensional fluid computational domain is established based on the spatial geometric topology of the junction of the outlets of the two waste heat boilers and the main pipe to obtain the real-time mass flow rate and density parameters of the two steam streams. Based on the principle of momentum conservation, the momentum flux ratio of the two steam streams at the junction node is calculated, and the aerodynamic stagnation interface surface formed by the collision of the two fluid streams is solved accordingly. Based on the spatial distribution of the aerodynamic stagnation interface surface in the three-dimensional fluid computation domain, the effective flow cross section and effective hydraulic diameter of the weak-side fluid after being squeezed by the strong-side fluid are calculated.

7. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 6, characterized in that, The step of generating the physical feature set of flow field distortion further includes: Using the aerodynamic stagnation interface surface as the initial boundary condition for shear mixing, and combining the Kelvin-Helmholtz instability characteristics induced by the momentum flux ratio, a spatial distribution model of the shear mixing layer extending along the parent tube axis is constructed. The physical spatial coordinates of the temperature sensor on the main pipe are projected into the spatial distribution model of the shear mixing layer, and the geometric deviation vector between the sensor probe position and the true energy centroid of the fluid is calculated. Combining the correction effect of the effective hydraulic diameter on flow measurement and the correction effect of the geometric deviation vector on enthalpy measurement, the physical feature set of flow field distortion is generated.

8. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 4, characterized in that, The steps of outputting the equivalent steam flow set and the equivalent thermodynamic state set further include: The asymmetric load correction factor is applied to the actual measured data of steam metering and thermodynamic state corresponding to the asymmetric operating section. An iterative reweighting strategy based on robust regression is used to suppress outliers, and the equivalent steam flow set and the equivalent thermodynamic state set are output.

9. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 1, characterized in that, The steps for outputting the reliable energy efficiency index set include: Using the equivalent steam flow set and the equivalent thermodynamic state set as input, a virtual main pipe thermodynamic state is generated based on the shared main pipe topology, so that the contributions of the two waste heat boilers to the turbine steam inlet and heating extraction steam can be allocated in the digital domain. Based on the thermal state of the virtual main pipe, the equivalent energy quantification results of the power generation output channel, the heat supply output channel and the unavailable loss channel are calculated to form an equivalent energy channel set. The overall plant thermal efficiency is calculated from the equivalent energy channel set, and the energy efficiency confidence characterization quantity is generated based on the reconstructed confidence interval, cross-channel consistency verification results and segment stability, and the confidence energy efficiency index set is output.

10. The data-driven intelligent energy efficiency analysis method for gas-fired steam turbine units according to claim 1, characterized in that, The steps for generating pseudo-high energy efficiency alarm and fluctuation suppression display parameters include: The set of reliable energy efficiency indicators is mapped to indicator points that can be subscribed to by the distributed control system, and a quality label corresponding to the calculation cycle is output. Based on the energy efficiency credibility representation quantity and the preset pseudo-high energy efficiency rule base, it is determined whether the current energy efficiency display is falsely high; When a false high energy efficiency is detected, an alarm signal and visual suppression parameters are generated, and operational recommendations for asymmetric heating regulation are output for operators' reference.

Images