A method for monitoring the net heat value of gas of a gas turbine combined cycle unit
By deploying data acquisition equipment and deep learning models in gas turbine combined cycle units, and dynamically adjusting the acquisition range and frequency, the error problem in monitoring the net calorific value of gas was solved, achieving accurate monitoring and adaptive capabilities, and improving operational economy and safety.
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
- 2026-06-05
AI Technical Summary
Existing methods for monitoring the net calorific value of gas in combined cycle gas turbine units are susceptible to sensor drift, insufficient representativeness of sampling points, or environmental interference. They are difficult to adapt to scenarios with varying fuel compositions or rapidly changing operating conditions. Furthermore, traditional methods cannot identify deviation trends in real time, leading to the accumulation of data errors and false alarms.
By acquiring direct calorific value data through acquisition devices deployed at preset monitoring points, and combining deep learning algorithms to build a simulation model, the acquisition range and frequency are dynamically adjusted. By using multi-source data verification and contribution analysis, key sensors are identified and calibrated, and targeted equipment adjustment suggestions are generated.
It enables accurate monitoring of the net calorific value of gas under complex operating conditions, reduces errors, improves data reliability and the adaptive capability of the monitoring system, reduces maintenance costs, and improves the economy and safety of unit operation.
Smart Images
Figure CN122149872A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data monitoring technology, specifically a method for monitoring the net calorific value of gas in a gas turbine combined cycle unit. Background Technology
[0002] Combined cycle gas turbine units, as a highly efficient power generation technology, are widely used in the power, energy, and industrial sectors. Their core principle is to generate electricity through a gas turbine and recover exhaust heat to drive a steam turbine, achieving cascaded energy utilization. The net calorific value (NCV) of the gas is a key parameter that directly affects combustion efficiency, emission control, and unit operational stability. In actual operation, the composition of the fuel may vary due to its source, transportation, or storage conditions, leading to fluctuations in NCV. Therefore, real-time and accurate monitoring of the NCV is fundamental to ensuring the efficient and safe operation of the unit.
[0003] Modern gas turbine combined cycle systems are typically equipped with multiple types of sensors, distributed at key locations such as the combustion chamber and exhaust pipes, forming a data acquisition network to provide multi-source data support for calorific value monitoring.
[0004] Existing methods for monitoring the net calorific value of natural gas still have shortcomings. Direct measurement is susceptible to sensor drift, insufficient representativeness of sampling points, or environmental interference, leading to data deviation. Indirect calculation relies on idealized models, which are difficult to adapt to scenarios with varying fuel composition or rapid changes in operating conditions, resulting in significant error accumulation. Traditional methods often use static thresholds or fixed calibration cycles, which cannot identify deviation trends in real time, nor can they dynamically adjust monitoring strategies according to the unit's operating mode, potentially leading to missed anomalies or false alarms. When calorific value data is abnormal, existing technologies struggle to accurately locate the key sensors or sampling points that cause the deviation. Summary of the Invention
[0005] (a) Technical problems to be solved In view of the above-mentioned shortcomings of the prior art, the present invention provides a method for monitoring the net calorific value of gas in a gas turbine combined cycle unit, which can effectively solve the problems of the prior art.
[0006] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a method for monitoring the net calorific value of gas in a gas turbine combined cycle unit, comprising the following steps: Step 1: Obtain direct calorific value measurement data of the gas by using acquisition equipment deployed at preset monitoring points; simultaneously acquire index measurement data of the gas turbine combustion chamber, and calculate the indirect calorific value data of the corresponding points based on the index measurement data; Step 2: Identify the initial deviation fluctuation trend between the indirect calorific value data and the direct calorific value data at each point, and determine whether the initial deviation fluctuation trend exceeds the allowable range based on the safety threshold of the current operating mode; the safety threshold range is dynamically set according to the different load conditions of the gas turbine combined cycle unit, the history of fuel type switching, and the multiple operating modes pre-divided by the ambient temperature range; the judgment process also includes a stability assessment of the initial deviation fluctuation trend. If the trend is within the threshold range but exhibits a continuous unidirectional drift characteristic, an early warning is triggered. Step 3: If the determination in Step 2 is negative, then based on the physical collection range of the heat value collection device and the index collection device at the abnormal point, and combined with the actual deviation coefficient obtained from historical data analysis, generate several verification collection instruction packages; among them, each instruction package contains detailed operation instructions for the collection direction, range and switching time of the device at the point in a specific working mode. Step 4: Construct a simulation pre-run model based on deep learning algorithms. This model takes the current operating mode of the unit, basic configuration parameters, and verification acquisition instruction package sequence as input, simulates the entire process of data acquisition and subsequent processing after executing different instruction packages, and outputs the highly correlated verification acquisition instruction package with the highest correlation to the current operating mode and optimization target. Step 5: Send the highly correlated verification and acquisition instruction packet to the corresponding sensing device, and control it to switch to the specified acquisition range for data re-acquisition within a predetermined time sequence; the predetermined time sequence is set according to the stability of the unit's operating conditions: when the unit is in steady-state operation, the switching and re-acquisition process is completed in a short time; when the unit is in a dynamic process of changing load, the data acquisition window is extended, and the re-acquisition data is subjected to moving average or dynamic filtering to remove noise introduced by transient changes in operating conditions; Step 6: Based on the re-collected data obtained in Step 5, the final deviation fluctuation trend is re-identified through the contribution analysis algorithm; Step 7: Mark the high-impact sensor data that leads to the final deviation fluctuation trend, and generate corresponding adjustment data for the operating settings of the collected equipment. Based on the high-impact sensor data category marked in Step 6 and the deviation characteristics it exhibits, back-map to the operating status and setting parameters of the collected equipment that generated the data, and generate targeted equipment operating setting adjustment data to guide maintenance personnel in calibrating or maintaining the monitoring equipment.
[0007] Furthermore, the index measurement data collected in step 1 includes fuel flow rate, air flow rate, exhaust temperature, exhaust pressure, and flue gas oxygen content; the indirect calorific value data is obtained by back-calculation through a preset thermodynamic model. The thermodynamic model establishes the heat balance equation or combustion efficiency equation of the gas turbine by taking fuel flow rate, air flow rate, exhaust temperature, pressure, and flue gas oxygen content as input variables, and based on the principles of energy conservation and combustion chemistry, iterative calculations are used to match the theoretical exhaust parameters output by the model with the actual measured values, so as to back-calculate the net calorific value of the gas as a key input variable.
[0008] Furthermore, the process of identifying the initial fluctuation deviation trend in step 2 is as follows: For the same monitoring point, within the same timestamp or sampling period, the difference between the indirect calorific value data and the direct calorific value data is calculated to obtain the instantaneous deviation value. In order to reduce instantaneous interference, the instantaneous deviation values of multiple consecutive sampling periods are processed by moving average or exponential weighted average to obtain the current deviation value sequence of the point. The time series analysis method is used to extract the trend of the current deviation value sequence. Linear regression is applied to fit the straight line of deviation value change over time. The slope of the line represents the overall direction and rate of deviation change. The discrete fluctuation amplitude of the deviation is quantified by calculating the standard deviation or root mean square value of the deviation value sequence within a certain time window. The change direction, rate, and fluctuation amplitude are combined to form a quantitative description of the initial deviation fluctuation trend. If the absolute value of the slope is greater than zero and continues for more than a preset duration, it is determined that there is a unidirectional drift trend. If the fluctuation amplitude exceeds the stable operation threshold corresponding to the current working mode, it is determined that there is an abnormal fluctuation trend. The initial deviation fluctuation trend is a comprehensive judgment result that includes drift characteristics and fluctuation characteristics.
[0009] Furthermore, the data acquisition equipment in step 1 includes a calorific value analyzer, a flow sensor, a temperature sensor, a pressure sensor, and a flue gas content sensor; in step 3, when generating the verification data acquisition instruction package, the switchable acquisition range of the data acquisition equipment includes: for the calorific value analyzer, different positions on the movable track designed for its sampling probe; for the flow, temperature, pressure, and flue gas content sensors, redundant measuring points installed at different radial or axial positions on the same pipeline as backups; the actual deviation coefficient is obtained by analyzing the statistical characteristics of data deviation under different acquisition orientations in historical data.
[0010] Furthermore, the process of generating the verification acquisition instruction packet in step 3 is as follows: A monitoring point information database is pre-established, which stores the physical acquisition range supported by the calorific value acquisition equipment and various index acquisition equipment deployed at each preset monitoring point. The acquisition range includes the movable spatial position of the equipment sampling probe, the installation coordinates of redundant measurement points, and the adjustable sampling frequency and integration time. A mapping relationship database between the acquisition range and different unit operating modes is established, and the recommended priority acquisition directions are defined under different loads, fuel types, or environmental conditions. Based on historical data from monitoring points, the deviation sequence between indirect calorific value data and direct calorific value data obtained under different historical time periods, collection locations, or settings is calculated. Statistical analysis is performed on this deviation sequence to calculate its mean, variance, and correlation coefficient with changes in collection location. The resulting actual deviation coefficient is used to quantify the reliability level of data within a specific collection range. For the identified points with abnormal initial deviation fluctuation trends, the mapping relationship library is queried to obtain all switchable alternative acquisition ranges for that point in the current working mode. Combined with the actual deviation coefficients corresponding to each alternative acquisition range, and with the goal of optimizing subsequent data quality, specific acquisition orientation parameters, start and stop times for range switching, and duration for maintaining stable acquisition after switching are configured for each alternative range, generating several independent verification acquisition instruction packages.
[0011] Furthermore, the process of constructing the simulation pre-run model in step 4 is as follows: Step 41: Construct a deep learning model architecture that includes an encoder, simulator, and evaluator; Step 42: Collect a historical dataset containing complete data sequences before and after executing different acquisition command packages under different working modes, the corresponding actual deviation fluctuation trend changes, and the final equipment adjustment effect; use the historical dataset to perform supervised training and reinforcement learning training on the simulation pre-run model, optimize the model parameters, minimize the error between the model's predicted output and the real result, and enable the model to learn to maximize the optimization objective function; Step 43: Deploy the converged training model to the online monitoring system; each time Step 4 is executed, call the model to simulate and evaluate the currently generated verification acquisition instruction package sequence; at the same time, use the actual execution results generated in Steps 5 to 7 as new training samples, and periodically perform incremental updates to the model to continuously adapt to changes in the unit equipment status and working mode.
[0012] Furthermore, in step 41, the encoder is used to encode the current operating mode, basic configuration parameters, and verification acquisition instruction package to be evaluated of the gas turbine combined cycle unit into a unified feature vector; the simulator is a multi-layer neural network jointly trained based on physical mechanisms and historical data, used to receive the feature vector, simulate the acquisition actions defined by the instruction package, and predict the reacquisition direct calorific value data and reacquisition index measurement data to be obtained; the evaluator is used to calculate the predicted reacquisition indirect calorific value based on the predicted data output by the simulator, and evaluate the estimated improvement in data quality after executing the instruction package and its correlation with the current operating mode based on a preset optimization objective function.
[0013] Furthermore, the contribution analysis process in step 6 employs a feature-ranking importance method to quantify the contribution of changes in each sensor data category to the fluctuation of the final deviation value. Data categories whose contribution exceeds a preset proportion of the total contribution are marked as high-impact data. The calculation formula is as follows: ; In the formula, Representing the The contribution of each sensor data category is calculated, with a value ranging from (0,1). A larger value indicates a greater impact of that sensor data on the final bias. This represents the total number of sampling points within the analysis time window. Represents a point in time The final deviation value, that is, the difference between the direct calorific value data and the indirect calorific value data, Representing the Each sensor data category at a given time point The measured value, Representing the Each sensor data category at a given time point The deviation of the measured value from its historical mean, Represents the final deviation value for the first The partial derivative of a single sensor data point reflects the sensitivity of that data point to changes in deviation. This represents the total number of sensor data categories, including fuel flow, air flow, exhaust temperature, pressure, and flue gas oxygen content. Represents the final deviation value for the first Partial derivatives of individual sensor data, Representing the Each sensor data category at a given time point The measured value, Representing the Each sensor data category at a given time point The deviation of the measured value from its historical mean.
[0014] Furthermore, the device operation setting adjustment data generated in step 7 includes: for sensors marked as high impact, the recommended calibration cycle reduction amount, zero drift compensation value, range adjustment parameters, installation position fine-tuning scheme, or sampling pipeline purging and maintenance plan.
[0015] Furthermore, after the equipment adjustment data generated in step 7 is applied to actual equipment maintenance, the monitoring process is restarted from step 1. The adjustment effect is verified using the new round of collected data, and the verification results are fed back to the simulation pre-run model in step 4 as training reference to continuously optimize the decision accuracy of the deep learning model.
[0016] (III) Beneficial Effects Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: 1. By integrating direct calorific value measurement and indirect calorific value data derived from multiple indices, a dual-source verification mechanism is constructed. Combined with deep learning simulation and verification of acquisition command optimization, the optimal acquisition location and range are dynamically selected to reduce errors introduced by fixed sensor positions or environmental interference. Through final deviation trend re-identification and contribution analysis, noise is further filtered out to ensure that the output data truly reflects the calorific value of the gas, providing a highly reliable input for unit control.
[0017] 2. By introducing dynamic judgment of safety thresholds driven by working mode and simulation pre-run model based on reinforcement learning, the monitoring process can adapt to complex operating conditions such as unit load changes and fuel switching. By generating and executing highly correlated verification acquisition instructions, the online optimization of monitoring strategy can be realized. Sudden deviations can be dealt with without manual intervention, improving decision accuracy and promoting the evolution of the monitoring system from passive response to proactive prediction.
[0018] 3. By marking high-impact sensor data and generating targeted equipment adjustment suggestions, it can accurately guide maintenance personnel to calibrate or repair key equipment, avoid blind and comprehensive overhaul, reduce maintenance costs and time, and form a sensor health status database in the long term, which can help preventive maintenance, reduce unplanned downtime, improve unit availability and overall operating economy, and provide data support for fuel management and emission control. Attached Figure Description
[0019] 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.
[0020] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a flowchart illustrating the simulation and pre-model construction process in this invention. Detailed Implementation
[0021] 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.
[0022] The present invention will be further described below with reference to embodiments. Example 1
[0023] This embodiment provides a method for monitoring the net calorific value of gas in a gas turbine combined cycle unit, such as... Figure 1 As shown, it includes the following steps: Step 1: Acquire direct calorific value measurement data of the gas by deploying data acquisition equipment at preset monitoring points. The data acquisition equipment includes a calorific value analyzer, flow sensor, temperature sensor, pressure sensor, and flue gas content sensor. Simultaneously acquire index measurement data of the gas turbine combustion chamber, and back-calculate the indirect calorific value data of the corresponding points based on the index measurement data. The index measurement data acquired in Step 1 includes fuel flow rate, air flow rate, exhaust temperature, exhaust pressure, and flue gas oxygen content. The indirect calorific value data is obtained by back-calculation using a preset thermodynamic model. The thermodynamic model establishes the heat balance equation or combustion efficiency equation of the gas turbine by using fuel flow rate, air flow rate, exhaust temperature, pressure, and flue gas oxygen content as input variables. Based on the principles of energy conservation and combustion chemistry, iterative calculations are used to match the theoretical exhaust parameters output by the model with the actual measured values, so as to back-calculate the net calorific value of the gas as a key input variable.
[0024] Step 2: Identify the initial deviation fluctuation trend between the indirect calorific value data and the direct calorific value data at each point, and determine whether the initial deviation fluctuation trend exceeds the allowable range based on the safety threshold of the current operating mode; the safety threshold range is dynamically set according to the different load conditions of the gas turbine combined cycle unit, the history of fuel type switching, and the multiple operating modes pre-divided by the ambient temperature range; the judgment process also includes a stability assessment of the initial deviation fluctuation trend. If the trend is within the threshold range but exhibits a continuous unidirectional drift characteristic, an early warning is triggered; the identification process of the initial fluctuation deviation trend is as follows: For the same monitoring point, within the same timestamp or sampling period, the difference between the indirect calorific value data and the direct calorific value data is calculated to obtain the instantaneous deviation value. In order to reduce instantaneous interference, the instantaneous deviation values of multiple consecutive sampling periods are processed by moving average or exponential weighted average to obtain the current deviation value sequence of the point. The time series analysis method is used to extract the trend of the current deviation value sequence. Linear regression is applied to fit the straight line of deviation value change over time. The slope of the line represents the overall direction and rate of deviation change. The discrete fluctuation amplitude of the deviation is quantified by calculating the standard deviation or root mean square value of the deviation value sequence within a certain time window. The change direction, rate, and fluctuation amplitude are combined to form a quantitative description of the initial deviation fluctuation trend. If the absolute value of the slope is greater than zero and continues for more than a preset duration, it is determined that there is a unidirectional drift trend. If the fluctuation amplitude exceeds the stable operation threshold corresponding to the current working mode, it is determined that there is an abnormal fluctuation trend. The initial deviation fluctuation trend is a comprehensive judgment result that includes drift characteristics and fluctuation characteristics.
[0025] Step 3: If the determination in Step 2 is negative, then based on the physical collection range of the calorific value collection device and the index collection device at the abnormal point, and combined with the actual deviation coefficient obtained from historical data analysis, several verification collection instruction packages are generated. Each instruction package contains detailed operation instructions for the collection orientation, range, and switching time of the device at that point in a specific working mode. When generating the verification collection instruction package, the switchable collection range of the collection device includes: for the calorific value analyzer, different positions on the movable track designed for its sampling probe; for the flow, temperature, pressure, and flue gas content sensors, redundant measuring points installed at different radial or axial positions on the same pipeline as backups. The actual deviation coefficient is obtained by analyzing the statistical characteristics of data deviation under different collection orientations in historical data.
[0026] The process of generating the verification acquisition instruction packet is as follows: A monitoring point information database is pre-established, which stores the physical acquisition range supported by the calorific value acquisition equipment and various index acquisition equipment deployed at each preset monitoring point. The acquisition range includes the movable spatial position of the equipment sampling probe, the installation coordinates of redundant measurement points, and the adjustable sampling frequency and integration time. A mapping relationship database between the acquisition range and different unit operating modes is established, and the recommended priority acquisition directions are defined under different loads, fuel types, or environmental conditions. Based on historical data from monitoring points, the deviation sequence between indirect calorific value data and direct calorific value data obtained under different historical time periods, collection locations, or settings is calculated. Statistical analysis is performed on this deviation sequence to calculate its mean, variance, and correlation coefficient with changes in collection location. The resulting actual deviation coefficient is used to quantify the reliability level of data within a specific collection range. For the identified points with abnormal initial deviation fluctuation trends, the mapping relationship library is queried to obtain all switchable alternative collection ranges for that point in the current working mode. Combined with the actual deviation coefficients corresponding to each alternative collection range, and with the goal of optimizing subsequent data quality, specific collection orientation parameters, start and stop time points for range switching, and duration of stable collection after switching are configured for each alternative range, generating several independent verification collection instruction packages. By pre-constructing a mapping library between the collection range and the working mode, and introducing actual deviation coefficients based on historical statistics, the system can automatically generate a series of parameterized verification instruction packages for specific abnormal points. This overcomes the shortcomings of traditional methods, which rely on human experience, have a single verification scheme, and are out of touch with the current operating conditions. It provides an evaluable optimization basis for subsequent deep learning-based simulation and pre-running, and systematically improves the adaptability and diagnostic accuracy of calorific value monitoring.
[0027] Step 4: Construct a simulation pre-run model based on deep learning algorithms. This model takes the current operating mode of the unit, basic configuration parameters, and verification acquisition instruction package sequence as input, simulates the entire process of data acquisition and subsequent processing after executing different instruction packages, and outputs the highly correlated verification acquisition instruction package with the highest correlation to the current operating mode and optimization target. Step 5: Send the highly correlated verification and acquisition instruction packet to the corresponding sensing device, and control it to switch to the specified acquisition range for data re-acquisition within a predetermined time sequence; the predetermined time sequence is set according to the stability of the unit's operating conditions: when the unit is in steady-state operation, the switching and re-acquisition process is completed in a short time; when the unit is in a dynamic process of changing load, the data acquisition window is extended, and the re-acquisition data is subjected to moving average or dynamic filtering to remove noise introduced by transient changes in operating conditions; Step 6: Based on the re-collected data obtained in Step 5, the final deviation fluctuation trend is re-identified through the contribution analysis algorithm; Step 7: Mark the high-impact sensor data that leads to the final deviation fluctuation trend, and generate corresponding adjustment data for the operating settings of the collected equipment. Based on the high-impact sensor data category marked in Step 6 and the deviation characteristics it exhibits, back-map to the operating status and setting parameters of the collected equipment that generated the data, and generate targeted equipment operating setting adjustment data to guide maintenance personnel in calibrating or maintaining the monitoring equipment.
[0028] The generated equipment operation setting adjustment data includes: for sensors marked as high-impact, the recommended calibration cycle reduction, zero-point drift compensation value, range adjustment parameters, installation position fine-tuning scheme, or sampling pipeline purging maintenance plan; for example, when the marked high-impact sensor data is fuel flow data, the generated adjustment data is a correction instruction for the opening of the fuel regulating valve; when the marked high-impact sensor data is exhaust temperature data, the generated adjustment data is a correction instruction for the inlet guide vane angle or burner fuel distribution; after the generated equipment adjustment data is applied to actual equipment maintenance, the monitoring process is restarted from step 1, and the adjustment effect is verified using the new round of collected data. The verification results are fed back as training data to the simulation pre-run model in step 4 as a training reference for continuously optimizing the decision accuracy of the deep learning model.
[0029] Compared with existing technologies, by integrating direct measurement and indirect calorific value data derived from combustion chamber parameters, and introducing initial deviation trend analysis and dynamic judgment of safety thresholds, early and accurate identification of calorific value monitoring anomalies is achieved. A deep learning-based simulation pre-running model is used to optimize and select multiple sets of verification acquisition commands, and drive the sensing equipment to perform adaptive range switching and data re-acquisition, thereby improving the targeting and reliability of data acquisition under complex operating conditions. Contribution analysis is used to identify high-impact sensing data and generate equipment adjustment suggestions, effectively overcoming the shortcomings of traditional methods that rely on fixed measuring points, are susceptible to local distortion, and have delayed maintenance, thus improving the economy and safety of unit operation. Example 2 At other levels, this embodiment also provides another optimization mechanism based on Embodiment 1, specifically a simulation pre-running model, such as... Figure 2 As shown, the process of constructing the simulation pre-model is as follows:
[0030] Step 41: Construct a deep learning model architecture including an encoder, a simulator, and an evaluator; the encoder is used to encode the current operating mode, basic configuration parameters, and verification acquisition instruction package to be evaluated of the gas turbine combined cycle unit into a unified feature vector; the simulator is a multi-layer neural network jointly trained based on physical mechanisms and historical data, used to receive the feature vector, simulate the acquisition actions defined by the instruction package, and predict the reacquisition direct calorific value data and reacquisition index measurement data to be obtained; the evaluator is used to calculate the predicted reacquisition indirect calorific value based on the predicted data output by the simulator, and evaluate the estimated improvement in data quality after executing the instruction package and its correlation with the current operating mode based on a preset optimization objective function.
[0031] Step 42: Collect a historical dataset containing complete data sequences before and after executing different acquisition command packages under different working modes, the corresponding actual deviation fluctuation trend changes, and the final equipment adjustment effect; use the historical dataset to perform supervised training and reinforcement learning training on the simulation pre-run model, optimize the model parameters, minimize the error between the model's predicted output and the real result, and enable the model to learn to maximize the optimization objective function.
[0032] Step 43: Deploy the converged training model to the online monitoring system; each time Step 4 is executed, call the model to simulate and evaluate the currently generated verification acquisition instruction package sequence; at the same time, use the actual execution results generated in Steps 5 to 7 as new training samples, and periodically perform incremental updates to the model to continuously adapt to changes in the unit equipment status and working mode. Example 3
[0033] In this embodiment, a formula for calculating contribution analysis is provided. Specifically, a feature-based importance ranking method is used to quantify the contribution of changes in each sensor data category to the fluctuation of the final deviation value. Data categories whose contribution exceeds a preset proportion of the total contribution are marked as high-impact data. The calculation formula is as follows: ; In the formula, Representing the The contribution of each sensor data category is calculated, with a value ranging from (0,1). A larger value indicates a greater impact of that sensor data on the final bias. This represents the total number of sampling points within the analysis time window. Represents a point in time The final deviation value, that is, the difference between the direct calorific value data and the indirect calorific value data, Representing the Each sensor data category at a given time point The measured value, Representing the Each sensor data category at a given time point The deviation of the measured value from its historical mean, Represents the final deviation value for the first The partial derivative of a single sensor data point reflects the sensitivity of that data point to changes in deviation. This represents the total number of sensor data categories, including fuel flow, air flow, exhaust temperature, pressure, and flue gas oxygen content. Represents the final deviation value for the first Partial derivatives of individual sensor data, Representing the Each sensor data category at a given time point The measured value, Representing the Each sensor data category at a given time point The deviation of the measured value from its historical mean; This embodiment quantifies the contribution of each sensor data point by calculating the degree of influence of the changes in each sensor data point on the final calorific value deviation. First, the measurement deviation of each sensor data point within the time window is calculated, and its sensitivity to the calorific value deviation is combined to obtain the local contribution of the data point at each time moment. Then, the contributions of all time points are accumulated and normalized to finally obtain the total contribution of each sensor data category. The higher the contribution, the greater the impact of the sensor data on the calorific value deviation, thus helping to identify key error sources. It not only considers the absolute deviation of the sensor data, but also its sensitivity to the calorific value deviation, which can more accurately reflect the influence weight under different operating conditions. By accumulating the calculation within the time window, it avoids the random error of single-point data, improves stability, and presents the contribution in the form of a ratio, which makes it easy to intuitively compare the degree of influence of different sensor data and optimize maintenance priorities.
[0034] In summary, this invention constructs a dual-path data source by simultaneously acquiring direct calorific value data and indirect calorific value data derived from key combustion chamber parameters. By comparing the deviation trends of the two, it can effectively identify single sensor failures or data drift issues, thus improving the overall reliability of calorific value monitoring. Furthermore, the introduction of a deep learning-driven simulation model enables the intelligent generation and selection of the optimal verification acquisition scheme without disrupting production. By executing optimized instructions for data re-acquisition and contribution analysis, it can accurately locate key sensors or measurement links causing deviations and generate specific equipment adjustment suggestions, significantly improving the accuracy and efficiency of maintenance. This invention fully considers the variable load, fuel, and environmental conditions of gas turbine combined cycle units, dynamically sets safety thresholds, generates verification instruction packages based on operating modes, and adjusts data acquisition timing according to operational stability. These designs enable the system to adapt to the actual operating state of the unit, provide timely warnings of potential risks, ensure operational safety, and provide a reliable data foundation for combustion optimization by ensuring the accuracy of calorific value measurement, thereby supporting the economical and efficient operation of the unit.
[0035] 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 monitoring the net calorific value of gas in a gas turbine combined cycle unit, characterized in that, Includes the following steps: Step 1: Obtain direct calorific value measurement data of natural gas by deploying acquisition equipment at preset monitoring points; The index measurement data of the gas turbine combustion chamber are collected synchronously, and the indirect calorific value data of the corresponding points are obtained by back-calculation based on the index measurement data. Step 2: Identify the initial deviation fluctuation trend between the indirect calorific value data and the direct calorific value data at each point, and determine whether the initial deviation fluctuation trend exceeds the allowable range based on the safety threshold of the current working mode; Step 3: If the determination in Step 2 is negative, then based on the physical collection range of the heat value collection device and the index collection device at the abnormal point, and combined with the actual deviation coefficient obtained from historical data analysis, generate several verification collection instruction packages. Step 4: Construct a simulation pre-run model based on deep learning algorithms. This model takes the current operating mode of the unit, basic configuration parameters, and verification acquisition instruction packet sequence as input, and outputs a highly correlated verification acquisition instruction packet that has the highest correlation with the current operating mode and optimization target. Step 5: Send the highly correlated verification and acquisition instruction packet to the sensing device at the corresponding location, and control it to switch to the specified acquisition range for data acquisition within the predetermined time sequence; Step 6: Based on the re-collected data obtained in Step 5, the final deviation fluctuation trend is re-identified through the contribution analysis algorithm; Step 7: Mark the sensor data with high impact coefficients that cause the final deviation fluctuation trend, and generate corresponding adjustment data for the operating settings of the acquired device.
2. The method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, The index measurement data collected in step 1 include fuel flow rate, air flow rate, exhaust temperature, exhaust pressure, and flue gas oxygen content; the indirect calorific value data is obtained by back-calculation using a preset thermodynamic model.
3. The method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, The process of identifying the initial fluctuation deviation trend in step 2 is as follows: For the same monitoring point, within the same timestamp or sampling period, the difference between the indirect calorific value data and the direct calorific value data is calculated to obtain the instantaneous deviation value. The instantaneous deviation values of multiple consecutive sampling periods are processed by moving average or exponential weighted average to obtain the current deviation value sequence of the point. The time series analysis method is used to extract the trend of the current deviation value sequence. Linear regression is applied to fit the straight line of deviation value change over time. The slope of the line represents the overall direction and rate of deviation change. The discrete fluctuation amplitude of the deviation is quantified by calculating the standard deviation or root mean square value of the deviation value sequence within a certain time window. The direction, rate, and amplitude of change are combined to form a quantitative description of the initial deviation fluctuation trend. If the absolute value of the slope is greater than zero and continues for more than a preset duration, it is determined that there is a unidirectional drift trend. If the fluctuation amplitude exceeds the stable operation threshold corresponding to the current working mode, it is determined that there is an abnormal fluctuation trend.
4. The method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, The data acquisition equipment in step 1 includes a calorific value analyzer, a flow sensor, a temperature sensor, a pressure sensor, and a flue gas content sensor. In step 3, when generating the verification data acquisition instruction package, the data acquisition range that the data acquisition equipment can switch between includes: for the calorific value analyzer, different positions on the movable track designed for its sampling probe; for the flow, temperature, pressure, and flue gas content sensors, redundant measuring points installed at different radial or axial positions on the same pipeline as backups.
5. The method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, The process of generating the verification acquisition instruction packet in step 3 is as follows: A database of monitoring points is pre-established, which stores the physical collection range supported by the calorific value acquisition equipment and various indicator acquisition equipment deployed at each preset monitoring point. A database of mapping relationships between collection range and different unit operating modes is established, and the recommended priority collection locations are defined under different loads, fuel types or environmental conditions. Based on historical data from monitoring points, the deviation sequence between indirect calorific value data and direct calorific value data obtained under different historical time periods, different collection locations or settings is calculated. The deviation sequence is statistically analyzed, and the actual deviation coefficient is obtained by combining the results. For the identified points with abnormal initial deviation fluctuation trends, the mapping relationship library is queried to obtain all switchable alternative acquisition ranges for that point in the current working mode. Combined with the actual deviation coefficients corresponding to each alternative acquisition range, and with the goal of optimizing subsequent data quality, specific acquisition orientation parameters, start and stop times for range switching, and duration for maintaining stable acquisition after switching are configured for each alternative range, generating several independent verification acquisition instruction packages.
6. The method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, The process of constructing the simulation pre-run model in step 4 is as follows: Step 41: Construct a deep learning model architecture that includes an encoder, simulator, and evaluator; Step 42: Collect a historical dataset containing complete data sequences before and after executing different acquisition command packages under different working modes, the corresponding actual deviation fluctuation trends, and the final equipment adjustment effects; use the historical dataset to perform supervised training and reinforcement learning training on the simulation pre-run model; Step 43: Deploy the converged training model to the online monitoring system.
7. The method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 6, characterized in that, The encoder in step 41 encodes the current operating mode, basic configuration parameters, and verification acquisition instruction package to be evaluated of the gas turbine combined cycle unit into a unified feature vector. The simulator is a multi-layer neural network jointly trained based on physical mechanisms and historical data. It receives the feature vector, simulates the acquisition actions defined by the instruction package, and predicts the reacquisition direct calorific value data and reacquisition index measurement data to be obtained. The evaluator calculates the predicted reacquisition indirect calorific value based on the predicted data output by the simulator, and evaluates the estimated improvement in data quality after executing the instruction package and its correlation with the current operating mode based on a preset optimization objective function.
8. The method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, The contribution analysis process in step 6 employs a feature-ranking importance method to quantify the contribution of changes in each sensor data category to the fluctuation of the final deviation value. Data categories whose contribution exceeds a preset proportion of the total contribution are marked as high-impact data. The calculation formula is as follows: ； In the formula, Representing the The contribution of each sensor data category This represents the total number of sampling points within the analysis time window. Represents a point in time The final deviation value, Representing the Each sensor data category at a given time point The measured value, Representing the Each sensor data category at a given time point The deviation of the measured value from its historical mean, Represents the final deviation value for the first Partial derivatives of individual sensor data, The total number of categories representing sensor data. Represents the final deviation value for the first Partial derivatives of individual sensor data, Representing the Each sensor data category at a given time point The measured value, Representing the Each sensor data category at a given time point The deviation of the measured value from its historical mean.
9. A method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, The device operation setting adjustment data generated in step 7 includes: for sensors marked as high impact, the recommended calibration cycle reduction, zero drift compensation value, range adjustment parameters, installation position fine-tuning scheme, or sampling pipeline purging and maintenance plan.
10. A method for monitoring the net calorific value of gas in a gas turbine combined cycle unit according to claim 1, characterized in that, After the equipment adjustment data generated in step 7 is applied to actual equipment maintenance, the monitoring process is restarted from step 1. The adjustment effect is verified using the new round of collected data, and the verification results are fed back to the simulation pre-run model in step 4 as training reference.