Method and system for on-line performance evaluation of combined cycle units
By constructing a reinforcement learning algorithm model in the combined cycle unit, the operating parameters and power generation data are analyzed in real time, which solves the problem that existing technologies cannot evaluate performance in real time. This enables early identification of performance degradation trends and provides optimization adjustments, thereby improving operational initiative and equipment lifespan.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA ENERGY CO LTD
- Filing Date
- 2026-03-18
- Publication Date
- 2026-06-19
Smart Images
Figure CN122241200A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data evaluation technology, specifically to a method and system for online performance evaluation of combined cycle units. Background Technology
[0002] Combined cycle power generation technology, due to its high efficiency, low emissions, and excellent operational flexibility, has become an important component of modern power systems, particularly suitable for grid peak shaving and serving as a core component of regional energy supply. A typical combined cycle unit consists of a gas turbine, a waste heat boiler, and a steam turbine. Its essence is energy cascade utilization: the high-temperature exhaust gas from the gas turbine generates electricity enters the waste heat boiler to produce steam, which then drives the steam turbine to generate electricity again. As the energy industry moves towards intelligent and refined operation and management, higher demands are placed on the real-time status perception, performance maintenance, and optimization of these high-value assets.
[0003] Existing technologies for performance evaluation of combined cycle units have the following main shortcomings: Traditional methods rely heavily on periodic performance tests or offline thermodynamic calculations, which are time-consuming, cannot reflect real-time operating conditions, and are difficult to detect transient or gradual performance degradation in a timely manner; Most online monitoring systems only issue threshold alarms and lack in-depth analysis of the complex relationships between multiple parameters within the combined cycle unit system, making it difficult to quickly and accurately trace the source to specific combustion or power generation operation settings when abnormal fluctuations occur in power generation; Existing systems generally focus on status description and post-event alarms, lacking models that can quantitatively predict the risk of future performance degradation based on historical data and operating mechanisms, and cannot provide parameter adjustment suggestions with operational guidance, leading to operators often only responding passively after problems become apparent, potentially missing the optimal maintenance window. Summary of the Invention
[0004] (a) Technical problems to be solved In view of the above-mentioned shortcomings of the prior art, the present invention provides a method and system for online performance evaluation of combined cycle units, which can effectively solve the problems of the prior art.
[0005] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, the present invention discloses an online performance evaluation method for a combined cycle unit, the combined cycle unit comprising at least one gas turbine and one steam turbine driven by the exhaust waste heat of the gas turbine, comprising the following steps: Step 1: Identify and acquire the operating setting parameters related to combustion of the gas turbine, the operating setting parameters related to power generation of the steam turbine, and the real-time power generation data of the unit within the preset monitoring period; Step 2: Based on the preset power generation fluctuation characteristic index, process the real-time power generation data, extract the power generation fluctuation characteristics within the current monitoring period, compare the power generation fluctuation characteristics with the preset normal fluctuation threshold, and determine whether there are abnormal power generation fluctuation characteristics; the normal fluctuation threshold is obtained based on the statistical learning of the unit's historical normal operation data, and is dynamically corrected according to the season and environmental benchmark parameters. Step 3: When the frequency or amplitude of the abnormal power generation fluctuation characteristics exceeds a preset abnormal threshold, extract the gas turbine combustion operation setting parameters and steam turbine power generation operation setting parameters that are time-correlated with the abnormal power generation fluctuation characteristics from the time series database, as an associated operation setting dataset; the time correlation refers to: extracting the historical time series data of the operation setting parameters of all gas turbines and steam turbines within the first preset time window before the occurrence of the abnormal power generation fluctuation characteristics, during the occurrence period, and the second preset time window thereafter; Step 4: Calculate the correlation between each operating setting parameter in the associated operating setting dataset and the abnormal power generation fluctuation characteristics, and filter out the operating setting parameters with a correlation higher than the preset correlation threshold, and mark them as key influencing factors; Step 5: Input the key influencing factors and their corresponding abnormal power generation fluctuation characteristics into the unit performance degradation prediction model pre-constructed by reinforcement learning algorithm. The model outputs the future performance degradation risk coefficient corresponding to each key influencing factor. The risk coefficient is used to quantify the probability and severity of the power generation fluctuation degradation that may occur if the current state of the operating settings continues. Step 6: Based on the future performance degradation risk coefficient, combined with the current operating conditions of the unit and equipment constraints, generate optimization adjustment parameter suggestions for key influencing factors, output an evaluation report containing the key influencing factors, their corresponding future performance degradation risk coefficients and optimization adjustment parameter suggestions, and trigger the corresponding early warning or control command interface.
[0006] Furthermore, the operating settings parameters in step 1 include: fuel flow valve opening, fuel distribution ratio, combustion chamber inlet temperature, compressor guide vane angle, and combustion mode; the operating settings parameters also include: main steam pressure, main steam temperature, condenser vacuum, feedwater temperature, and circulating water flow rate.
[0007] Furthermore, the extraction process of power generation fluctuation characteristics in step 2 includes: Step 21: Filter and denoise the real-time power generation data to eliminate measurement noise and instantaneous interference. Standardize or normalize the denoised power data to eliminate the influence of dimensions and adapt it to subsequent feature extraction. Step 22: Based on the preset first set of power generation fluctuation characteristic indicators, calculate the statistical characteristics of the power data in the current monitoring period in the time domain; Step 23: Perform a fast Fourier transform or wavelet transform on the real-time power generation data to convert the time-domain signal to the frequency domain; extract features in the frequency domain based on the preset second set of power generation fluctuation characteristic indicators; Step 24: The time-domain features extracted in Step 22 and the frequency-domain features extracted in Step 23 are fused to form a feature vector, which is the digital expression representing the power generation fluctuation characteristics within the current monitoring period.
[0008] Furthermore, the first set of characteristic indicators in step 22 includes: the standard deviation of the power sequence, the absolute average deviation, the maximum value and root mean square value of the power difference between adjacent sampling points, and the number of times the power change rate exceeds a preset threshold; the second set of characteristic indicators in step 23 includes at least: the power spectral density integral value, the main fluctuation frequency components and their amplitudes in the predefined low-frequency band, mid-frequency band and high-frequency band, and the ratio of energy in a specific frequency band to total energy.
[0009] Furthermore, the correlation degree in step 4 is calculated using grey relational analysis, which quantifies the similarity between the time-series curves of each operating setting parameter and the time-series curves of abnormal power generation fluctuations. The calculation formula is as follows: ; In the formula, Representing the The correlation between a certain operating setting parameter and the characteristics of abnormal power generation fluctuations is calculated, with a value range of (0,1). A larger value indicates a stronger correlation. Represents the total number of sampling points. Representing the Standardized values of abnormal power generation fluctuation characteristics at each sampling point Representing the The first running setting parameter is in the... Standardized values of each sampling point This represents the resolution coefficient, used to adjust the calculation sensitivity. Its value ranges from (0,1), with a default value of 0.5. This represents the minimum difference between two levels, that is, the minimum absolute difference between all parameters and the reference sequence at all sampling points. This represents the maximum difference between the two levels, that is, the maximum absolute difference between all parameters and the reference sequence at all sampling points.
[0010] Furthermore, the process of constructing the unit performance degradation prediction model in step 5 is as follows: Collect historical gas turbine combustion operation parameters, steam turbine power generation operation parameters, power generation time series data, and corresponding unit performance degradation event labels; perform data preprocessing, including outlier removal, missing value imputation, time alignment, and standardization. Construct a state space, where each state consists of current operating parameters and power generation fluctuation characteristics; define an action space, including operating parameter adjustment strategies, such as fuel flow adjustment and guide vane angle optimization; set a reward function, based on power generation stability, equipment safety constraints, and energy efficiency indicators, to obtain the benefit or penalty after each adjustment; A deep reinforcement learning algorithm is used to train an agent to explore the optimal control strategy in a simulated environment or historical data. During training, the agent selects actions based on the current state, the environment returns a new state and reward, and the policy network is updated to maximize the long-term cumulative reward. The model's ability to predict the adjustment of operating parameters and the deterioration trend of power generation is optimized by combining offline training and online fine-tuning. Model validation is conducted until it passes validation. The trained model is then integrated into the online monitoring system to receive operational data in real time and output the risk coefficient of deterioration. Model validation methods include: using cross-validation to evaluate the model's prediction accuracy and ensure its generalization ability on unseen data; correcting the model output based on expert experience to improve interpretability and engineering applicability; and periodically using new data to incrementally train the model to adapt to changes in the unit's operating status.
[0011] Furthermore, the future performance degradation risk coefficient in step 5 includes: the expected percentage increase in power generation fluctuation within a specified future time period, the expected frequency of abnormal fluctuation events, and the estimated total power generation loss.
[0012] Furthermore, the process of generating the optimization adjustment parameter suggestions in step 6 is as follows: taking the minimization of the future performance degradation risk coefficient as the core optimization objective, and using the safe operation boundary, physical constraints, and preset steady-state operation range of the unit equipment as constraints, a target optimization model is constructed; the target optimization model uses a solution algorithm to calculate one or more sets of parameter adjustment values that can effectively reduce the risk coefficient to a preset threshold in the parameter space composed of the key influencing factors, and outputs the set of parameter adjustment values with the best overall benefit as the final optimization adjustment parameter suggestion.
[0013] Secondly, this invention discloses an online performance evaluation system for combined cycle power units, comprising: The data acquisition module is used to monitor and acquire the operating data of the combined cycle unit in real time. The data includes: operating setting parameters related to combustion of the gas turbine, operating setting parameters related to power generation of the steam turbine, and real-time power generation data of the unit within a preset monitoring period. The feature initial judgment module is used to process real-time power generation data based on preset power generation fluctuation feature indicators, extract power generation fluctuation features within the current monitoring period, and compare the power generation fluctuation features with preset normal fluctuation thresholds to determine whether there are abnormal power generation fluctuation features. The anomaly extraction module is used to extract gas turbine combustion operation setting parameters and steam turbine power generation operation setting parameters that are temporally associated with the abnormal power generation fluctuation characteristics from the time series database when the frequency or amplitude of abnormal power generation fluctuation characteristics exceeds the preset anomaly threshold, forming an associated operation setting dataset. The factor screening module is used to calculate the correlation between each operating setting parameter in the associated operating setting dataset and the abnormal power generation fluctuation characteristics, and to screen out operating setting parameters with a correlation higher than the preset correlation threshold, marking them as key influencing factors. The risk prediction module is used to pre-build a unit performance degradation prediction model through reinforcement learning algorithm. The key influencing factors and their corresponding abnormal power generation fluctuation characteristics are input into the degradation prediction model. The model outputs the future performance degradation risk coefficient corresponding to each key influencing factor. The risk coefficient is used to quantify the probability and severity of the potential degradation of power generation fluctuations if the current operating setting continues. The adjustment generation module is used to generate optimization adjustment parameter suggestions for key influencing factors based on the future performance degradation risk coefficient, combined with the current operating conditions of the unit and equipment safety constraints retrieved from the knowledge base, through a reverse optimization algorithm. The evaluation output module is used to integrate and output a structured evaluation report containing key influencing factors, their corresponding risk coefficients for future performance degradation, and suggestions for optimization and adjustment parameters. It is also equipped with an interface to trigger corresponding early warning prompts or send adjustment instructions to the unit control system.
[0014] Furthermore, the data acquisition module and the feature preliminary judgment module are interconnected via a wireless network; the anomaly extraction module, the feature preliminary judgment module, and the risk prediction module are interconnected via a wireless network; the factor screening module and the risk prediction module are interconnected via a wireless network; and the risk prediction module, the adjustment generation module, and the evaluation output module are interconnected via a wireless network.
[0015] (III) Beneficial Effects Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: 1. By continuously collecting key operating settings and power generation data of gas turbines and steam turbines through preset cycles, and automatically identifying abnormal power generation fluctuations based on feature matching indicators, continuous status assessment is carried out. Once abnormal features are detected, the analysis process is immediately started, which greatly shortens the time from the occurrence of an anomaly to its recognition.
[0016] 2. By using reinforcement learning algorithms to construct a deterioration prediction model, the selected key operating settings and current abnormal features are used as inputs, and the output is a quantified future deterioration risk coefficient. Based on the current state, the model predicts the degree of fluctuation and deterioration that will result from the continued improper operation of specific operating parameters. This allows operators to prioritize high-risk items based on the risk coefficient and proactively adjust the operating mode before significant performance degradation or failure occurs, thereby avoiding unplanned downtime, reducing efficiency losses, and extending equipment life. Attached Figure Description
[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0018] Figure 1 This is a flowchart illustrating the online performance evaluation method for combined cycle units in this invention. Figure 2 This is a schematic diagram of the framework of the online performance evaluation system for combined cycle units in this invention.
[0019] The labels in the diagram represent: 1. Data acquisition module; 2. Preliminary feature judgment module; 3. Anomaly extraction module; 4. Factor screening module; 5. Risk prediction module; 6. Adjustment and generation module; 7. Evaluation output module. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0021] The present invention will be further described below with reference to embodiments. Example 1
[0022] The online performance evaluation method for combined cycle units in this embodiment, such as... Figure 1 As shown, a combined cycle unit includes at least one gas turbine and one steam turbine driven by the exhaust waste heat of the gas turbine, and includes the following steps: Step 1: Identify and acquire the operating setting parameters related to combustion of the gas turbine, the operating setting parameters related to power generation of the steam turbine, and the real-time power generation data of the unit within the preset monitoring period; the operating setting parameters include: fuel flow valve opening, fuel distribution ratio, combustion chamber inlet temperature, compressor guide vane angle, and combustion mode; the operating setting parameters include: main steam pressure, main steam temperature, condenser vacuum, feedwater temperature, and circulating water flow rate.
[0023] Step 2: Based on preset power generation fluctuation characteristic indicators, process real-time power generation data, extract power generation fluctuation characteristics within the current monitoring period, and compare these characteristics with preset normal fluctuation thresholds to determine if any abnormal power generation fluctuation characteristics exist. The normal fluctuation thresholds are obtained through statistical learning based on historical normal operating data of the unit and are dynamically adjusted according to seasonal and environmental baseline parameters. The process of extracting power generation fluctuation characteristics includes: Step 21: Filter and denoise the real-time power generation data to eliminate measurement noise and instantaneous interference. Standardize or normalize the denoised power data to eliminate the influence of dimensions and adapt it to subsequent feature extraction. Step 22: Based on the preset first set of power generation fluctuation characteristic indicators, calculate the statistical characteristics of the power data in the current monitoring period in the time domain; the first set of characteristic indicators includes: the standard deviation of the power sequence, the absolute mean deviation, the maximum value and root mean square value of the power difference between adjacent sampling points, and the number of times the power change rate exceeds the preset threshold; the second set of characteristic indicators in Step 23 includes at least: the power spectral density integral value, the main fluctuation frequency components and their amplitudes in the predefined low frequency band, mid frequency band and high frequency band, and the ratio of energy in a specific frequency band to the total energy; Step 23: Perform Fast Fourier Transform or Wavelet Transform on the real-time power generation data to convert the time-domain signal to the frequency domain; extract features in the frequency domain based on the preset second set of power generation fluctuation characteristic indicators; Step 24: The time-domain features extracted in Step 22 and the frequency-domain features extracted in Step 23 are fused to form a feature vector, which is the digital expression representing the power generation fluctuation characteristics within the current monitoring period.
[0024] Step 3: When the frequency or amplitude of abnormal power generation fluctuations exceeds the preset abnormal threshold, extract the gas turbine combustion operation setting parameters and steam turbine power generation operation setting parameters that are time-correlated with the abnormal power generation fluctuations from the time series database as the associated operation setting dataset; time correlation refers to: extracting the historical time series data of all gas turbine and steam turbine operation setting parameters within the first preset time window before the occurrence of the abnormal power generation fluctuations, during the occurrence period, and the second preset time window afterward.
[0025] Step 4: Calculate the correlation between each operating setting parameter in the associated operating setting dataset and the abnormal power generation fluctuation characteristics, and filter out the operating setting parameters with a correlation higher than the preset correlation threshold, marking them as key influencing factors.
[0026] Step 5: Input the key influencing factors and their corresponding abnormal power generation fluctuation characteristics into the unit performance degradation prediction model pre-built using reinforcement learning algorithms. The model outputs a future performance degradation risk coefficient corresponding to each key influencing factor. The risk coefficient is used to quantify the probability and severity of power generation fluctuation degradation that may result if the current operating settings continue. The construction process of the unit performance degradation prediction model is as follows: Collect historical gas turbine combustion operation parameters, steam turbine power generation operation parameters, power generation time series data, and corresponding unit performance degradation event labels; perform data preprocessing, including outlier removal, missing value imputation, time alignment, and standardization. Construct a state space, where each state consists of current operating parameters and power generation fluctuation characteristics; define an action space, including operating parameter adjustment strategies, such as fuel flow adjustment and guide vane angle optimization; set a reward function, based on power generation stability, equipment safety constraints, and energy efficiency indicators, to obtain the benefit or penalty after each adjustment; A deep reinforcement learning algorithm is used to train an agent to explore the optimal control strategy in a simulated environment or historical data. During training, the agent selects actions based on the current state, the environment returns a new state and reward, and the policy network is updated to maximize the long-term cumulative reward. The model's ability to predict the adjustment of operating parameters and the deterioration trend of power generation is optimized by combining offline training and online fine-tuning. Model validation is conducted until it passes validation. The trained model is then integrated into the online monitoring system to receive operational data in real time and output a degradation risk coefficient. The future performance degradation risk coefficient includes: the expected percentage increase in power generation fluctuation within a specified future time period, the expected frequency of abnormal fluctuation events, and the estimated total power generation loss. Model validation methods include: using cross-validation to evaluate the model's prediction accuracy and ensure its generalization ability on unseen data; revising the model output based on expert experience to improve interpretability and engineering applicability; and periodically using new data for incremental training to adapt to changes in unit operating status. By combining the operating parameters of gas turbines and steam turbines with the characteristics of power generation fluctuations, the dynamic correlation between parameters is quantified through improved grey relational analysis, overcoming the problem of insufficient sensitivity of single parameter monitoring. By adopting a deep reinforcement learning framework and autonomously exploring the optimal control strategy through a reward mechanism, it can identify potential deterioration trends earlier and adapt to changes in unit status compared to traditional threshold alarms or static models.
[0027] Step 6: Based on the future performance degradation risk coefficient, combined with the current operating conditions of the unit and equipment constraints, generate optimization adjustment parameter suggestions for key influencing factors, output an evaluation report containing key influencing factors, their corresponding future performance degradation risk coefficients, and optimization adjustment parameter suggestions, and trigger the corresponding early warning or control command interface; use the actual operating setting data of the current monitoring period, the actual power generation fluctuation characteristics, and the feedback data of the effect after subsequent adjustment measures as new training samples, and periodically or triggered incrementally learn and update the degradation prediction model pre-built by the reinforcement learning algorithm to achieve adaptive optimization of the model.
[0028] The process of generating optimized adjustment parameter recommendations is as follows: with minimizing the risk coefficient of future performance degradation as the core optimization objective, and with the safe operation boundary, physical constraints, and preset steady-state operation range of the unit equipment as constraints, a target optimization model is constructed; the target optimization model uses a solution algorithm to back-optimize and calculate one or more sets of parameter adjustment values that can effectively reduce the risk coefficient to a preset threshold in the parameter space composed of key influencing factors, and outputs the set of parameter adjustment values with the best overall benefit as the final optimized adjustment parameter recommendation. With the core objective of minimizing the risk coefficient of future performance degradation, and under the premise of meeting the unit's safety boundaries, physical constraints, and steady-state operating range, a solution algorithm performs reverse optimization in the parameter space of key influencing factors to calculate the parameter adjustment values that can effectively reduce the risk coefficient. The set with the best overall benefit is then selected as the final recommendation. Its advantages lie in its ability to systematically balance performance optimization and operational safety, improve the accuracy and reliability of adjustments through quantitative optimization, and take into account multiple constraints to ensure the feasibility and robustness of the recommendations in actual operation. Example 2 At other levels, this embodiment also provides another optimization mechanism based on Embodiment 1, specifically an online performance evaluation system for combined cycle units, such as... Figure 2 As shown, it includes:
[0029] Data acquisition module 1 is used to monitor and acquire the operating data of the combined cycle unit in real time. The data includes: operating setting parameters related to combustion of the gas turbine, operating setting parameters related to power generation of the steam turbine, and real-time power generation data of the unit within a preset monitoring period. Feature Preliminary Judgment Module 2 is used to process real-time power generation data based on preset power generation fluctuation feature indicators, extract power generation fluctuation features within the current monitoring period, and compare the power generation fluctuation features with preset normal fluctuation thresholds to determine whether there are abnormal power generation fluctuation features. The anomaly extraction module 3 is used to extract gas turbine combustion operation setting parameters and steam turbine power generation operation setting parameters that are temporally associated with the abnormal power generation fluctuation characteristics from the time series database when the frequency or amplitude of the abnormal power generation fluctuation characteristics exceeds the preset anomaly threshold, and form an associated operation setting dataset. Factor screening module 4 is used to calculate the correlation between each operating setting parameter in the associated operating setting dataset and the abnormal power generation fluctuation characteristics, and to screen out operating setting parameters with a correlation higher than the preset correlation threshold and mark them as key influencing factors. Risk prediction module 5 is used to predict the deterioration of unit performance in advance through reinforcement learning algorithm. The key influencing factors and their corresponding abnormal power generation fluctuation characteristics are input into the deterioration prediction model. The model outputs the future performance deterioration risk coefficient corresponding to each key influencing factor. The risk coefficient is used to quantify the probability and severity of the deterioration of power generation fluctuation if the current state of the operating setting continues. The adjustment generation module 6 is used to generate optimization adjustment parameter suggestions for key influencing factors based on the risk coefficient of future performance degradation, combined with the current operating conditions of the unit and equipment safety constraints retrieved from the knowledge base, through a reverse optimization algorithm. The evaluation output module 7 is used to integrate and output a structured evaluation report containing key influencing factors, their corresponding risk coefficients for future performance degradation, and suggestions for optimization and adjustment parameters. It is also equipped with an interface to trigger corresponding early warning prompts or send adjustment instructions to the unit control system.
[0030] Data acquisition module 1 and feature preliminary judgment module 2 are interconnected via a wireless network. Anomaly extraction module 3, feature preliminary judgment module 2, and risk prediction module 5 are interconnected via a wireless network. Factor screening module 4 and risk prediction module 5 are interconnected via a wireless network. Risk prediction module 5, adjustment generation module 6, and evaluation output module 7 are interconnected via a wireless network. Example 3
[0031] In this embodiment, a method for calculating correlation degree is provided. Grey relational analysis is used to quantify the similarity between the time-series curves of various operating setting parameters and the time-series curves of abnormal power generation fluctuations. The calculation formula is as follows: ; In the formula, Representing the The correlation between a certain operating setting parameter and the characteristics of abnormal power generation fluctuations is calculated, with a value range of (0,1). A larger value indicates a stronger correlation. Represents the total number of sampling points. Representing the Standardized values of abnormal power generation fluctuation characteristics at each sampling point Representing the The first running setting parameter is in the... Standardized values of each sampling point This represents the resolution coefficient, used to adjust the calculation sensitivity. Its value ranges from (0,1), with a default value of 0.5. This represents the minimum difference between two levels, that is, the minimum absolute difference between all parameters and the reference sequence at all sampling points. This represents the maximum difference between the two levels, that is, the maximum absolute difference between all parameters and the reference sequence at all sampling points; The correlation calculation formula is based on grey relational analysis. It judges the correlation by quantifying the geometric similarity between the abnormal power generation fluctuation characteristic sequence and the sequence of various operating setting parameters. The data is standardized to eliminate the influence of dimensions. Then, the absolute difference between the two sequences at each sampling point is calculated, and a resolution coefficient is introduced to adjust the sensitivity to extreme values. Finally, the comprehensive correlation is obtained by weighted averaging the correlation coefficients of all sampling points. If the time series change trend of a certain operating setting is highly synchronized with the abnormal power generation fluctuation, that is, the difference fluctuation is small, the correlation is close to 1, indicating that the parameter may be the key factor causing the anomaly. Grey relational analysis does not rely on large data samples and strict normal distribution assumptions. It is suitable for non-stationary, small-sample time series data commonly found in combined cycle units. Through two-level range normalization, it focuses more on the overall trend matching degree between parameters and abnormal fluctuations, rather than relying solely on numerical linear correlation. It can capture lag or nonlinear correlations that are easily ignored by existing technologies. The resolution coefficient allows for dynamic adjustment of the sensitivity to local mutations, avoiding misjudgments caused by noise interference in traditional methods. It is especially suitable for scenarios with transient disturbances such as unstable combustion in gas turbines.
[0032] In summary, this invention, through periodic dynamic monitoring and feature matching, can promptly capture abnormal power generation fluctuations hidden in conventional data, effectively identify early signs of degradation that may be overlooked by traditional methods, not only trace the key operating settings that cause performance fluctuations, but also use a reinforcement learning-based prediction model to quantitatively assess the potential risk of the current operating state to future performance degradation. The system automatically generates targeted adjustment parameter suggestions, providing data-driven decision support for operators. This guides them to take preventative measures and make adjustments before problems escalate, avoiding unplanned shutdowns or deep performance degradation. Ultimately, this ensures the long-term stable operation of the unit and improves overall power generation efficiency and equipment lifespan.
[0033] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for online performance evaluation of a combined cycle unit, wherein the combined cycle unit comprises at least one gas turbine and one steam turbine driven by the exhaust waste heat of the gas turbine, characterized in that, Includes the following steps: Step 1: Identify and acquire the operating setting parameters related to combustion of the gas turbine, the operating setting parameters related to power generation of the steam turbine, and the real-time power generation data of the unit within the preset monitoring period; Step 2: Based on the preset power generation fluctuation characteristic index, process the real-time power generation data, extract the power generation fluctuation characteristics within the current monitoring period, compare the power generation fluctuation characteristics with the preset normal fluctuation threshold, and determine whether there are abnormal power generation fluctuation characteristics. Step 3: When the frequency or amplitude of abnormal power generation fluctuations exceeds the preset abnormal threshold, extract the gas turbine combustion operation setting parameters and steam turbine power generation operation setting parameters that are time-related to the abnormal power generation fluctuations from the time series database as the associated operation setting dataset. Step 4: Calculate the correlation between each operating setting parameter in the associated operating setting dataset and the abnormal power generation fluctuation characteristics, and filter out the operating setting parameters with a correlation higher than the preset correlation threshold, and mark them as key influencing factors; Step 5: Input the key influencing factors and their corresponding abnormal power generation fluctuation characteristics into the unit performance degradation prediction model pre-built using reinforcement learning algorithm. The model outputs the future performance degradation risk coefficient corresponding to each key influencing factor. Step 6: Based on the risk coefficient of future performance degradation, and combined with the current operating conditions of the unit and equipment constraints, generate optimization adjustment parameter suggestions for key influencing factors.
2. The online performance evaluation method for combined cycle units according to claim 1, characterized in that, The operating settings parameters in step 1 include: fuel flow valve opening, fuel distribution ratio, combustion chamber inlet temperature, compressor guide vane angle, and combustion mode; the operating settings parameters include: main steam pressure, main steam temperature, condenser vacuum, feedwater temperature, and circulating water flow rate.
3. The online performance evaluation method for combined cycle units according to claim 1, characterized in that, The extraction process of power generation fluctuation characteristics in step 2 includes: Step 21: Filter and denoise the real-time power generation data, and standardize or normalize the denoised power data. Step 22: Based on the preset first set of power generation fluctuation characteristic indicators, calculate the statistical characteristics of the power data in the current monitoring period in the time domain; Step 23: Perform a fast Fourier transform or wavelet transform on the real-time power generation data to convert the time-domain signal to the frequency domain; extract features in the frequency domain based on the preset second set of power generation fluctuation characteristic indicators; Step 24: Fuse the time-domain features extracted in Step 22 with the frequency-domain features extracted in Step 23 to form a feature vector.
4. The online performance evaluation method for combined cycle units according to claim 3, characterized in that, The first set of characteristic indicators in step 22 includes: the standard deviation of the power sequence, the absolute average deviation, the maximum value and root mean square value of the power difference between adjacent sampling points, and the number of times the power change rate exceeds a preset threshold; the second set of characteristic indicators in step 23 includes at least: the power spectral density integral value, the main fluctuation frequency components and their amplitudes in the predefined low-frequency band, mid-frequency band and high-frequency band, and the ratio of energy in a specific frequency band to total energy.
5. The online performance evaluation method for combined cycle units according to claim 1, characterized in that, The correlation degree in step 4 is calculated using grey relational analysis, which quantifies the similarity between the time-series curves of each operating setting parameter and the time-series curves of abnormal power generation fluctuations. The calculation formula is as follows: ； In the formula, Representing the The correlation between operating setting parameters and abnormal power generation fluctuation characteristics Represents the total number of sampling points. Representing the Standardized values of abnormal power generation fluctuation characteristics at each sampling point Representing the The first running setting parameter is in the... Standardized values of each sampling point Represents the resolution coefficient. Represents the minimum difference between two levels. This represents the maximum difference between the two levels.
6. The online performance evaluation method for combined cycle units according to claim 1, characterized in that, The process of constructing the unit performance degradation prediction model in step 5 is as follows: Collect historical gas turbine combustion operating parameters, steam turbine power generation operating parameters, power generation time series data, and corresponding unit performance degradation event tags; Construct a state space, where each state consists of current operating parameters and power generation fluctuation characteristics; define an action space, including operating parameter adjustment strategies, such as fuel flow adjustment and guide vane angle optimization; set a reward function, based on power generation stability, equipment safety constraints, and energy efficiency indicators, to obtain the benefit or penalty after each adjustment; A deep reinforcement learning algorithm is used to train an agent to explore the optimal control strategy in a simulated environment or historical data. During training, the agent selects actions based on the current state, the environment returns a new state and reward, and the policy network is updated to maximize long-term cumulative rewards; Model validation is performed until it passes validation. The trained model is then integrated into the online monitoring system to receive operational data in real time and output the risk coefficient of deterioration.
7. The online performance evaluation method for combined cycle units according to claim 1, characterized in that, The future performance degradation risk coefficient in step 5 includes: the expected percentage increase in power generation fluctuation within a specified future time period, the expected frequency of abnormal fluctuation events, and the estimated total power generation loss.
8. The online performance evaluation method for combined cycle units according to claim 1, characterized in that, The process of generating the optimization adjustment parameter suggestions in step 6 is as follows: taking the minimization of the future performance degradation risk coefficient as the core optimization objective, and using the safe operation boundary, physical constraints, and preset steady-state operation range of the unit equipment as constraints, a target optimization model is constructed; the target optimization model uses a solution algorithm to calculate one or more sets of parameter adjustment values that can effectively reduce the risk coefficient to a preset threshold in the parameter space composed of the key influencing factors, and outputs the set of parameter adjustment values with the best overall benefit as the final optimization adjustment parameter suggestion.
9. A combined cycle unit performance online evaluation system, said system being an implementation system based on the combined cycle unit performance online evaluation method according to any one of claims 1-8, characterized in that, include: The data acquisition module is used to monitor and acquire the operating data of the combined cycle unit and the real-time power generation data of the unit in real time. The feature initial judgment module is used to process real-time power generation data based on preset power generation fluctuation feature indicators, extract power generation fluctuation features within the current monitoring period, and compare the power generation fluctuation features with preset normal fluctuation thresholds to determine whether there are abnormal power generation fluctuation features. The anomaly extraction module is used to extract gas turbine combustion operation setting parameters and steam turbine power generation operation setting parameters that are temporally associated with the abnormal power generation fluctuation characteristics from the time series database when the frequency or amplitude of abnormal power generation fluctuation characteristics exceeds the preset anomaly threshold, forming an associated operation setting dataset. The factor screening module is used to calculate the correlation between each operating setting parameter in the associated operating setting dataset and the abnormal power generation fluctuation characteristics, and to screen out operating setting parameters with a correlation higher than the preset correlation threshold, marking them as key influencing factors. The risk prediction module is used to pre-build a unit performance degradation prediction model through reinforcement learning algorithms. The key influencing factors and their corresponding abnormal power generation fluctuation characteristics are input into the degradation prediction model, and the model outputs the future performance degradation risk coefficient corresponding to each key influencing factor. The adjustment generation module is used to generate optimization adjustment parameter suggestions for key influencing factors based on the future performance degradation risk coefficient through a reverse optimization algorithm; The evaluation output module is used to integrate and output a structured evaluation report.
10. The online performance evaluation system for combined cycle units according to claim 9, characterized in that, The data acquisition module and the feature preliminary judgment module are interconnected via a wireless network. The anomaly extraction module, the feature preliminary judgment module, and the risk prediction module are interconnected via a wireless network. The factor screening module and the risk prediction module are interconnected via a wireless network. The risk prediction module, the adjustment generation module, and the evaluation output module are interconnected via a wireless network.