A method for correcting performance parameters of a steam turbine to suppress dynamic interference

Abstract

The application discloses a steam turbine performance parameter correction method for inhibiting dynamic interference, relates to a steam turbine performance parameter correction method, and aims at solving the problem that the correction result is greatly deviated and it is difficult to meet the variable working condition requirement in the prior art based on the quasi-steady state assumption. The application comprises the following steps: collecting steam turbine operation thermodynamic parameters to obtain unit basic performance values; adopting a deformable convolutional neural network to build a steam turbine variable working condition normal mode prediction model; inputting the operation thermodynamic parameters into the normal mode prediction model to output initial steam turbine performance parameters; utilizing error comparison to obtain an error actual value; setting an error threshold value to determine whether the error actual value exceeds the error threshold value; if yes, updating the normal mode prediction model to obtain a correction prediction model, inputting the collected operation thermodynamic parameters into the correction prediction model again to obtain corrected steam turbine performance parameters; otherwise, the initial steam turbine performance parameters do not need to be corrected. The correction result is within a reasonable range.

Description

Technical Field

[0001] This invention relates to a method for correcting the performance parameters of a steam turbine. Background Technology

[0002] Currently, most thermal power plants rely on static design parameters or typical steady-state operating condition data as evaluation benchmarks for the performance management of steam turbines, lacking dynamic simulation and online real-time monitoring methods that are adapted to all operating conditions and the entire process. Under complex dynamic operating scenarios such as unit load changes and deep peak shaving, the interstage pressure, steam temperature, enthalpy drop and other thermodynamic parameters of the steam turbine flow path will deviate significantly from the design values.

[0003] Existing performance parameter calculation methods cannot quantify the degree of unit performance degradation caused by issues such as flow path scaling, component wear, and valve leakage in a timely and accurate manner, resulting in a lack of reliable data support for equipment operation and maintenance and fault diagnosis. Furthermore, current technologies for analyzing the correlation of turbine operating parameters mostly rely on empirical formulas or local measurement data, making it difficult to comprehensively analyze the coupled impact of factors such as vacuum deterioration and reheat steam parameter fluctuations on unit heat consumption and output power. This makes it impossible to accurately pinpoint the root causes of energy efficiency losses, severely restricting the overall energy efficiency improvement of turbine generator units.

[0004] Furthermore, during actual operation, steam turbines are subject to dynamic factors such as unit volume space, pipeline heat storage, and medium flow leakage, which can cause irregular and large fluctuations in thermodynamic parameters and performance indicators. Existing thermodynamic performance correction methods based on quasi-steady-state assumptions cannot effectively suppress dynamic disturbances and correct parameters, resulting in large deviations in the correction results. This makes it difficult to meet the actual needs of accurate performance evaluation, online monitoring, and early warning of steam turbines under varying operating conditions. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that existing thermodynamic performance correction methods based on quasi-steady-state assumptions have large deviations in the correction results and are difficult to meet the requirements of variable operating conditions. This invention proposes a method for correcting turbine performance parameters to suppress dynamic disturbances.

[0006] The method for correcting turbine performance parameters to suppress dynamic disturbances according to the present invention includes the following steps:

[0007] The operating thermodynamic parameters of the steam turbine are collected, including main steam pressure, main steam temperature, back pressure, stage inlet and outlet enthalpy, extraction steam pressure, and regenerator outlet water temperature.

[0008] Based on the collected operating thermodynamic parameters, four performance indicators were calculated: turbine regulation and throttling loss efficiency, stage relative internal efficiency, regenerator terminal temperature difference, and condenser terminal temperature difference. These four performance indicators are collectively referred to as the unit's basic performance values.

[0009] A deformable convolutional neural network is used, with the collected operating thermodynamic parameters as model input and the unit's basic performance values ​​as model output, to construct the mapping relationship between the operating thermodynamic parameters and the unit's basic performance values, and to build a turbine variable operating condition constant mode prediction model.

[0010] The collected operating thermodynamic parameters are input into the normal mode prediction model, and the initial turbine performance parameters are output.

[0011] The actual error value is obtained by comparing the initial turbine performance parameters with the basic performance values ​​of the unit.

[0012] An error threshold is set, and it is determined whether the actual error value exceeds the error threshold. If so, the constant pattern prediction model is updated using incremental learning with feature transfer and weight fine-tuning to obtain a corrected prediction model. The collected operating thermodynamic parameters are then input into the corrected prediction model to obtain the corrected turbine performance parameters. Otherwise, the initial turbine performance parameters output by the constant pattern prediction model do not need to be corrected.

[0013] Furthermore, the efficiency of adjusting the throttling loss is calculated using the enthalpy drop formula;

[0014] The enthalpy drop formula is as follows:

[0015]

[0016] in, To adjust the efficiency of throttling loss; This represents the enthalpy of water vapor before the valve. This represents the actual enthalpy of water vapor after the valve. This represents the enthalpy of water vapor before pressure drop under isentropic conditions.

[0017] Furthermore, the formula for calculating the relative internal efficiency of the group is as follows:

[0018]

[0019] in, The relative internal efficiency of the level group; This represents the enthalpy of water vapor at the inlet position before the stage; This represents the actual enthalpy of water vapor at the stage outlet. The enthalpy of water vapor at the outlet of the stage under isentropic conditions is given.

[0020] Furthermore, the formula for calculating the terminal temperature difference of the regenerator is as follows:

[0021]

[0022] in, For the temperature difference at the regenerator terminals; To correspond to the extraction steam pressure The saturation temperature below; This refers to the outlet water temperature of the regenerator.

[0023] Furthermore, the formula for calculating the condenser terminal temperature difference is as follows:

[0024]

[0025] in, For condenser terminal differential; This refers to the condenser outlet water temperature.

[0026] Furthermore, the deformable convolutional neural network adaptively expands the feature extraction range by adding directional parameters to the convolution kernel elements, thereby extracting nonlinear operating data features of the steam turbine under varying operating conditions.

[0027] Furthermore, the specific formula for the constant pattern prediction model is as follows:

[0028]

[0029] in, Output for steam turbine and auxiliary equipment; To adjust the efficiency of throttling loss; The relative internal efficiency of the level group; For the temperature difference at the regenerator terminals; For condenser terminal differential; This is a stateful constant-pattern prediction model built using a deformable convolutional neural network. This refers to the main steam pressure of the steam turbine. Main steam pressure and temperature; Back pressure; This is the model function that corresponds to adjusting the efficiency of throttling loss; This is the model function for the relative internal efficiency of the corresponding class group; This is the model function corresponding to the temperature difference at the regenerator terminals; This is the model function corresponding to the condenser terminal temperature difference.

[0030] Compared with the prior art, the present invention has the following advantages:

[0031] This invention provides a method for correcting turbine performance parameters to suppress dynamic disturbances. Existing unit thermodynamic performance calculations based on quasi-steady-state assumptions are not applicable to online monitoring, and the complex and variable on-site operating environment of the unit makes it impossible to use preset rules to dynamically eliminate or correct thermodynamic performance. This invention, by employing advanced artificial intelligence technology and establishing a performance index prediction model (constant pattern prediction model) based on deformable convolutional neural networks, can effectively suppress the dynamic influences caused by unit volume space, pipeline heat storage, flow leakage, etc. Overall, based on the actual operating parameters of the unit (thermal parameters such as pressure, temperature, flow rate, and load), this invention can realize real-time calculation and analysis of turbine unit throttling losses, relative internal efficiency, regenerator terminal temperature difference, and condenser terminal temperature difference. It can also identify indicators that deviate significantly from design parameters, reminding operators to make adjustments, and guiding maintenance personnel to understand unit performance and formulate reasonable maintenance plans. Attached Figure Description

[0032] Figure 1 This is a schematic flowchart of a method for correcting turbine performance parameters to suppress dynamic disturbances, as described in Specific Implementation Method 1.

[0033] Figure 2 This is a schematic diagram of the enthalpy drop principle in the second specific implementation method;

[0034] Figure 3 This is a schematic diagram of the relative internal efficiency of the intermediate group in the third specific implementation method;

[0035] Figure 4 This is a schematic diagram of the regenerator in Specific Implementation Method Four;

[0036] Figure 5 This is a schematic diagram comparing the existing convolutional neural network (left) and the deformable convolutional neural network (right) in Specific Implementation Method Six;

[0037] Figure 6 This is a schematic diagram illustrating the anomaly detection principle of the normal pattern prediction model in Implementation Method Seven.

[0038] Figure 7 This is a schematic diagram illustrating the dynamic impact of relative internal efficiency in the seventh specific implementation method;

[0039] Figure 8 This is a schematic diagram of the dynamic elimination principle in the seventh specific implementation method;

[0040] Figure 9 This is a schematic diagram illustrating the dynamic performance evaluation and model update principle in Implementation Method Seven. Detailed Implementation

[0041] Specific Implementation Method 1: Combination Figure 1This embodiment describes a method for correcting turbine performance parameters to suppress dynamic disturbances, which includes the following steps:

[0042] The operating thermodynamic parameters of the steam turbine are collected, including main steam pressure, main steam temperature, back pressure, stage inlet and outlet enthalpy, extraction steam pressure, and regenerator outlet water temperature.

[0043] Based on the collected operating thermodynamic parameters, four performance indicators were calculated: turbine regulation and throttling loss efficiency, stage relative internal efficiency, regenerator terminal temperature difference, and condenser terminal temperature difference. These four performance indicators are collectively referred to as the unit's basic performance values.

[0044] A deformable convolutional neural network is used, with the collected operating thermodynamic parameters as model input and the unit's basic performance values ​​as model output, to construct the mapping relationship between the operating thermodynamic parameters and the unit's basic performance values, and to build a turbine variable operating condition constant mode prediction model.

[0045] The collected operating thermodynamic parameters are input into the normal mode prediction model, and the initial turbine performance parameters are output.

[0046] The actual error value is obtained by comparing the initial turbine performance parameters with the basic performance values ​​of the unit.

[0047] An error threshold is set, and it is determined whether the actual error value exceeds the error threshold. If so, the constant pattern prediction model is updated using incremental learning with feature transfer and weight fine-tuning to obtain a corrected prediction model. The collected operating thermodynamic parameters are then input into the corrected prediction model to obtain the corrected turbine performance parameters. Otherwise, the initial turbine performance parameters output by the constant pattern prediction model do not need to be corrected.

[0048] In this embodiment, four core performance indicators are calculated simultaneously based on thermal parameters to achieve full-dimensional performance quantification of the steam turbine. A constant-mode prediction model is constructed based on deformable convolutional neural networks, which can offset dynamic interference caused by unit volume, pipeline heat storage, and leakage, solving the problems of large fluctuations and high deviations in traditional quasi-steady-state calculations. An error threshold discrimination mechanism is set up, and the model is automatically updated through incremental learning when the error exceeds the limit, continuously ensuring long-term correction accuracy. The output is accurate performance parameters after eliminating dynamic interference, supporting unit anomaly identification, operation adjustment, and maintenance planning, and adapting to online monitoring scenarios of variable load and deep peak shaving.

[0049] Specific Implementation Method Two: Combination Figure 2 This embodiment further defines the method for correcting turbine performance parameters to suppress dynamic disturbances as described in Specific Embodiment 1. In this embodiment, the efficiency for adjusting throttling loss is calculated using the enthalpy drop formula.

[0050] The enthalpy drop formula is as follows:

[0051]

[0052] in, To adjust the efficiency of throttling loss; This represents the enthalpy of water vapor before the valve. This represents the actual enthalpy of water vapor after the valve. This represents the enthalpy of water vapor before pressure drop under isentropic conditions.

[0053] In this embodiment, throttling loss refers to the energy loss caused by changes in flow velocity and friction when the fluid passes through the regulating valve. Since the pressure downstream of the valve is unmeasurable, enthalpy drop is generally used. The principle is as follows: Figure 2 As shown, by using the ratio of the actual enthalpy value before and after the valve to the isentropic enthalpy value, the throttling loss is quantified, eliminating the need to rely on the difficult-to-measure downstream pressure and reducing the requirements for the arrangement of on-site measurement points. The enthalpy drop ratio directly reflects the energy loss caused by friction and flow rate changes in the regulating valve, accurately quantifying the degree of valve deterioration and providing a quantitative basis for valve leakage fault diagnosis.

[0054] Specific Implementation Method Three: Combination Figure 3 This embodiment further defines the method for correcting turbine performance parameters to suppress dynamic disturbances described in Specific Embodiment 1. In this embodiment, the formula for calculating the relative internal efficiency of the stage group is:

[0055]

[0056] in, The relative internal efficiency of the level group; This represents the enthalpy of water vapor at the inlet position before the stage; This represents the actual enthalpy of water vapor at the stage outlet. The enthalpy of water vapor at the outlet of the stage under isentropic conditions is given.

[0057] In this embodiment, the relative internal efficiency of the turbine stage is... This reflects the work done by water vapor in the flow path of the steam turbine. It is the ratio of the actual enthalpy drop of water vapor to the ideal isentropic enthalpy drop, and its definition is as follows: Figure 3 As shown, the ratio of actual enthalpy drop to ideal isentropic enthalpy drop characterizes the flow-through work capacity, accurately reflecting the efficiency decline caused by flow-through scaling and blade wear; it directly quantifies the energy efficiency loss of the turbine's flow-through components, quickly locates the root cause of insufficient unit output, and supports the judgment of when to overhaul the flow-through components.

[0058] Detailed Implementation Method Four: Combination Figure 4This embodiment further defines the method for correcting turbine performance parameters to suppress dynamic disturbances described in Specific Embodiment 1. In this embodiment, the calculation formula for the regenerator terminal temperature difference is:

[0059]

[0060] in, For the temperature difference at the regenerator terminals; To correspond to the extraction steam pressure The saturation temperature below; This refers to the outlet water temperature of the regenerator.

[0061] In this embodiment, the thermal efficiency and heat transfer effect of the regenerator are mainly evaluated by monitoring the terminal temperature difference (TTD). The principle is as follows: Figure 4 As shown, the heat exchange performance is evaluated by using the difference between the extraction steam saturation temperature and the outlet water temperature. The calculation logic is simple and the real-time performance is strong. It can quickly identify abnormalities such as heat exchange deterioration and tube blockage in the regenerator, guide the cleaning of the regenerator and the optimization of the condensate system, and improve the economic efficiency of the unit's regenerator cycle.

[0062] Specific Implementation Method Five: This implementation method further defines the turbine performance parameter correction method for suppressing dynamic disturbances described in Specific Implementation Method One. In this implementation method, the calculation formula for the condenser terminal temperature difference is:

[0063]

[0064] in, For condenser terminal differential; This refers to the condenser outlet water temperature.

[0065] In this embodiment, the condenser terminal temperature difference It represents the temperature difference between the cooling medium and the condensing steam, and is usually used to evaluate the heat exchange capacity of the condenser. The calculation method of the condenser end difference is consistent with that of the regenerator end difference calculation method. The temperature difference calculation logic is uniformly adopted and reused with the regenerator end difference algorithm, which simplifies the overall program development. It reflects the heat exchange capacity of the condenser in real time, quickly warns of vacuum deterioration and abnormal circulating water problems, and reduces the heat loss of the unit.

[0066] Specific Implementation Method Six: Combination Figure 5 This embodiment further defines the method for correcting turbine performance parameters to suppress dynamic disturbances described in Specific Embodiment 1. In this embodiment, the deformable convolutional neural network adaptively expands the feature extraction range by adding directional parameters to the convolution kernel elements, thereby extracting nonlinear operating data features of the turbine under varying operating conditions.

[0067] In this embodiment, considering the limitations of actual system measurement points and the requirement for computational accuracy, a high-precision constant-mode model of the steam turbine is established through deep learning, and the prediction results are used to accurately assess the performance changes of the unit. To meet the actual accuracy and real-time requirements, a deformable convolutional network (DCN) is selected as the steam turbine performance prediction model. DCN adaptively adds a direction parameter to each element of the convolution kernel, allowing the convolution kernel to expand to a larger range during training, which can significantly improve the model's feature extraction capability for nonlinear and variable operating condition data of the steam turbine. The convolution kernel can adaptively add a direction offset parameter to expand the feature extraction range and adapt to the strong nonlinear operating data of the steam turbine under variable operating conditions. Compared with existing convolutional networks, it significantly improves the fitting accuracy of thermodynamic parameters and performance indicators under dynamic operating conditions and strengthens the ability to suppress dynamic interference. The model is lightweight and has good real-time performance, meeting the online real-time calculation requirements of power plants.

[0068] Specific implementation method seven: combination Figures 6 to 9 This embodiment further defines the method for correcting turbine performance parameters to suppress dynamic disturbances described in Specific Embodiment 1. In this embodiment, the specific formula of the constant mode prediction model is:

[0069]

[0070] in, Output for steam turbine and auxiliary equipment; To adjust the efficiency of throttling loss; The relative internal efficiency of the level group; For the temperature difference at the regenerator terminals; For condenser terminal differential; This is a stateful constant-pattern prediction model built using a deformable convolutional neural network. This refers to the main steam pressure of the steam turbine. Main steam pressure and temperature; Back pressure; This is the model function that corresponds to adjusting the efficiency of throttling loss; This is the model function for the relative internal efficiency of the corresponding class group; This is the model function corresponding to the temperature difference at the regenerator terminals; This is the model function corresponding to the condenser terminal temperature difference.

[0071] In this embodiment, any output of a system can be described as a function of its input and state. When the system state remains unchanged, the system output can be accurately represented by the input. When the system state remains unchanged, the deviation between the model output and the actual output is within a certain range. When the deviation exceeds a certain threshold, it can be considered that the turbine state has changed (usually due to degradation or failure). Therefore, degradation detection can be simultaneously achieved using a constant-mode model. The main principle is as follows: Figure 6 As shown in Figure 7, the performance evaluation method for eliminating dynamic interference under varying operating conditions is as follows. Due to dynamic influences such as unit volume space, pipeline heat storage, and flow leakage, the actual operating characteristics of the steam turbine exhibit wide-range and nonlinear features. The calculation results using actual data from a certain unit are shown in Figure 7. It can be seen that the relative internal efficiency value fluctuates drastically and irregularly, and there are significant differences between load increases and decreases. In the process of establishing a performance index prediction model based on DCN, the network output value continuously calculates the error between the maximum and minimum measured values ​​under this operating condition and searches for the position of minimum error, i.e., the position closest to the theoretical law, as shown in Figure 7. Figure 8 As shown. Therefore, when the deviation of the established turbine constant-mode model is within the required range, its output can be directly used as the four performance indicators of the turbine after dynamic elimination; when the output deviation of the established turbine constant-mode model exceeds the tolerable range, it is necessary to re-import new data and update the model using incremental learning based on feature transfer and weight fine-tuning to obtain a new performance evaluation model that meets the requirements. The principle of this process is as follows: Figure 9 As shown; in summary, deformable convolution and incremental learning can be used to accurately calculate the performance indicators of steam turbines to eliminate dynamic disturbances, and at the same time, performance degradation can be detected. By establishing a unified mapping model for multiple performance indicators, a single network can simultaneously output four performance parameters, simplifying the modeling process; the minimum error output is selected within the measured index range to effectively filter out irregular fluctuations under rising and falling loads, thus completing the elimination of dynamic disturbances; after the error exceeds the limit, the model is updated through feature transfer and weight fine-tuning incremental learning, without the need for complete retraining, reducing data computing power consumption; relying on model deviation, automatic early warning of unit performance degradation and faults is achieved, taking into account both dynamic correction and online anomaly monitoring functions.

[0072] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for correcting turbine performance parameters to suppress dynamic disturbances, characterized in that, Includes the following steps: The operating thermodynamic parameters of the steam turbine are collected, including main steam pressure, main steam temperature, back pressure, stage inlet and outlet enthalpy, extraction steam pressure, and regenerator outlet water temperature. Based on the collected operating thermodynamic parameters, four performance indicators were calculated: turbine regulation and throttling loss efficiency, stage relative internal efficiency, regenerator terminal temperature difference, and condenser terminal temperature difference. These four performance indicators are collectively referred to as the unit's basic performance values. A deformable convolutional neural network is used, with the collected operating thermodynamic parameters as model input and the unit's basic performance values ​​as model output, to construct the mapping relationship between the operating thermodynamic parameters and the unit's basic performance values, and to build a turbine variable operating condition constant mode prediction model. The collected operating thermodynamic parameters are input into the normal mode prediction model, and the initial turbine performance parameters are output. The actual error value is obtained by comparing the initial turbine performance parameters with the basic performance values ​​of the unit. Set an error threshold and determine whether the actual error value exceeds the error threshold. If so, the constant pattern prediction model is updated using incremental learning with feature transfer and weight fine-tuning to obtain the corrected prediction model. The collected operating thermodynamic parameters are then input into the corrected prediction model to obtain the corrected turbine performance parameters. Otherwise, the initial turbine performance parameters output by the constant pattern prediction model do not need to be corrected.

2. The method for correcting turbine performance parameters to suppress dynamic disturbances according to claim 1, characterized in that, The efficiency of adjusting throttling loss is calculated using the enthalpy drop formula; The enthalpy drop formula is as follows: in, To adjust the efficiency of throttling loss; This represents the enthalpy of water vapor before the valve. This represents the actual enthalpy of water vapor after the valve. This represents the enthalpy of water vapor before pressure drop under isentropic conditions.

3. The method for correcting turbine performance parameters to suppress dynamic disturbances according to claim 1, characterized in that, The formula for calculating the relative internal efficiency of the group is: in, The relative internal efficiency of the level group; This represents the enthalpy of water vapor at the inlet position before the stage; This represents the actual enthalpy of water vapor at the stage outlet. The enthalpy of water vapor at the outlet of the stage under isentropic conditions is given.

4. The method for correcting turbine performance parameters to suppress dynamic disturbances according to claim 1, characterized in that, The formula for calculating the terminal temperature difference of the regenerator is: in, For the temperature difference at the regenerator terminals; To correspond to the extraction steam pressure The saturation temperature below; This refers to the outlet water temperature of the regenerator.

5. The method for correcting turbine performance parameters to suppress dynamic disturbances according to claim 4, characterized in that, The formula for calculating the condenser terminal temperature difference is: in, For condenser terminal differential; This refers to the condenser outlet water temperature.

6. The method for correcting turbine performance parameters to suppress dynamic disturbances according to claim 1, characterized in that, The deformable convolutional neural network described above adaptively expands the feature extraction range by adding directional parameters to the convolution kernel elements, thereby extracting nonlinear operating data features of steam turbines under varying operating conditions.

7. The method for correcting turbine performance parameters to suppress dynamic disturbances according to claim 1, characterized in that, The specific formula for the constant pattern prediction model is as follows: in, Output for steam turbine and auxiliary equipment; To adjust the efficiency of throttling loss; The relative internal efficiency of the level group; For the temperature difference at the regenerator terminals; For condenser terminal differential; This is a stateful constant-pattern prediction model built using a deformable convolutional neural network. This refers to the main steam pressure of the steam turbine. Main steam pressure and temperature; Back pressure; This is the model function that corresponds to adjusting the efficiency of throttling loss; This is the model function for the relative internal efficiency of the corresponding class group; This is the model function corresponding to the temperature difference at the regenerator terminals; This is the model function corresponding to the condenser terminal temperature difference.

Images