System and Method for Dynamic Parameter Compensation and Optimization of Turbine Controller Based on Physics-Informed Neural Networks (PINN)

Abstract

The present invention aims to provide a turbine control system and method capable of improving the output efficiency of a gas turbine while stably maintaining the control performance of an existing Functional Block Diagram (FBD)-based controller, even when the operating characteristics of the turbine change due to aging of the gas turbine, load fluctuations, and changes in ambient conditions. Furthermore, the embodiments disclosed herein aim to provide a turbine control system and method capable of estimating thermodynamic state variables, such as enthalpy and efficiency inside the turbine, which are difficult to measure directly, using a Physical Law-Compliant Artificial Neural Network (PINN), and calculating the output deviation between the current operating state and the theoretical optimal operating state based on the estimated state information.

Description

Technology Field

[0001] The present invention relates to gas turbine control technology used in weapon systems, and more specifically, to a turbine control system and method that improves the operating efficiency and stability of a gas turbine by estimating the thermodynamic state inside the turbine using a physical law-compliant artificial neural network (PINN) and dynamically compensating the control parameters of an existing functional block diagram (FBD)-based turbine controller based on the estimated state information. Background Technology

[0002] A gas turbine is a core component of power generation facilities that utilizes high-temperature, high-pressure gas generated by the combustion of fuel to convert thermal energy into mechanical rotational energy, thereby driving a generator. Such a gas turbine may consist of a compressor, combustor, turbine, and generator, and precise control based on operating conditions is required to ensure stable output and reduce fuel consumption.

[0003] Conventional gas turbine control systems are generally implemented using control logic based on Function Block Diagrams (FBDs). For example, conventional control systems receive operating data such as turbine pressure, temperature, flow rate, rotational speed, and generator output, and control the operating state of the gas turbine by generating control commands for guide vane valves, fuel valves, or other drive devices using a PID controller.

[0004] However, conventional FBD-based control methods often maintain control parameters, such as PID gains, time constants, and correction coefficients, at values ​​set during the initial commissioning or tuning phase. Consequently, there is a problem in that it is difficult to actively respond to changed operating conditions using only fixed control parameters, even if the turbine's dynamic or thermodynamic characteristics change due to equipment aging from long-term operation, load fluctuations, changes in ambient temperature and pressure, or changes in fuel composition.

[0005] Furthermore, thermodynamic state variables within the gas turbine, such as compressor efficiency, combustor state, enthalpy, change in entropy, and turbine efficiency, are difficult to measure directly or require separate, expensive sensors and complex instrumentation configurations. Consequently, conventional control systems rely on partial operational data measurable from external sensors to perform control, making it difficult to accurately determine the extent of output loss or efficiency degradation in the current operating state compared to the theoretically optimal operating state.

[0006] Meanwhile, recent attempts have been made to apply data-driven artificial intelligence models to improve the control performance of power generation facilities. However, general data-driven AI models merely generate output values ​​based on the statistical trends of training data and often fail to adequately reflect physical laws that must be applied to actual turbine systems, such as the law of energy conservation, the law of increasing entropy, and thermodynamic boundary conditions. In such cases, there is a risk that the AI ​​model may generate non-physical state values ​​or control commands that are unacceptable in the actual system, posing a problem that makes it difficult to apply to power generation facilities requiring high reliability and safety.

[0007] Therefore, a new turbine control technology is required that can estimate internal thermodynamic states, which are difficult to measure directly, based on gas turbine operating data, while also correcting the parameters of existing controllers in real time within a range consistent with physical laws. Prior art literature

[0008] Korean Patent Registration No. 10-1451837 (October 10, 2014) The problem to be solved

[0009] The embodiments disclosed in this disclosure aim to provide a turbine control system and method capable of improving the output efficiency of a gas turbine while stably maintaining the control performance of an existing functional block diagram (FBD)-based controller, even when the operating characteristics of the turbine change due to aging of the gas turbine, load fluctuations, and changes in ambient conditions.

[0010] Furthermore, the embodiments disclosed in this disclosure provide a turbine control system and method capable of estimating thermodynamic state variables, such as enthalpy and efficiency inside a turbine, which are difficult to measure directly, using a physical law-compliant artificial neural network (PINN), and calculating the output deviation between the current operating state and the theoretical optimal operating state based on the estimated state information.

[0011] Furthermore, the embodiments disclosed in this disclosure aim to provide a turbine control system and method capable of real-time control compensation while maintaining the structure of an existing control system by generating control correction parameters including PID gains or feedforward correction values ​​based on calculated output deviations and dynamically reflecting them in existing FBD control logic.

[0012] Furthermore, the embodiments disclosed in this disclosure provide a turbine control system and method capable of suppressing the generation of physically unacceptable state estimates or control commands by utilizing an artificial intelligence model that reflects the law of conservation of energy, entropy change conditions, and input / output boundary conditions.

[0013] Furthermore, the embodiments disclosed in this disclosure aim to provide a turbine control system and method capable of securing the operational stability and reliability required of a power generation facility by evaluating the output reliability of an artificial intelligence model and the safety range of control correction parameters, and by stopping the correction operation and returning to the existing control logic when reliability is low or an abnormal condition occurs.

[0014] Meanwhile, the technical problems of the present invention are not limited to those mentioned above, and other unmentioned technical problems will be clearly understood by a person skilled in the art from the description below. means of solving the problem

[0015] A turbine control system based on a physical law-compliant artificial neural network for achieving the technical task described above may include: a data collection and preprocessing unit (110) for collecting and preprocessing operating data of a gas turbine according to the present disclosure; a virtual sensor unit (120) for estimating thermodynamic state variables inside a turbine using a physical law-compliant artificial neural network (PINN) learned based on a loss function including physical laws, with the preprocessed operating data as input; an output deviation calculation unit (130) for calculating a theoretical output value under current operating conditions based on the estimated thermodynamic state variables and calculating an output deviation between the theoretical output value and the actual output value; a correction parameter generation unit (140) for generating control correction parameters to correct control parameters of an existing turbine controller based on the output deviation; a safety constraint and reliability evaluation unit (150) for evaluating whether the control correction parameters are applicable; and an FBD dynamic compensation module (160) for controlling the gas turbine by reflecting the control correction parameters in a functional block diagram (FBD)-based control logic. Effects of the invention

[0016] According to the aforementioned means for solving the problem of the present disclosure, thermodynamic state variables that are difficult to measure directly, such as enthalpy and efficiency inside a gas turbine, can be estimated in real time by using a physical law-compliant artificial neural network (PINN). Accordingly, compared to conventional control methods that rely simply on external sensor values, there is an effect of being able to determine the current operating state of the turbine more precisely.

[0017] Furthermore, according to the aforementioned means for solving the problem of the present disclosure, the deviation between the theoretical output and the actual output under current operating conditions is calculated based on estimated internal turbine state information, and control correction parameters including PID gains or feedforward correction values ​​are generated based thereon. Accordingly, the effect is to improve the operating efficiency of the gas turbine by correcting in real time the output loss caused by turbine aging, load fluctuations, and changes in ambient conditions.

[0018] Furthermore, according to the aforementioned means for solving the problem of the present disclosure, only a correction signal can be dynamically injected while maintaining the existing Functional Block Diagram (FBD)-based control logic without replacing it. Accordingly, control performance can be improved while maintaining the stability of the existing control system, and there is an effect of excellent applicability to existing power generation facilities.

[0019] Furthermore, according to the aforementioned means for solving the problem of the present disclosure, by utilizing a physical law-compliant artificial neural network that reflects the law of conservation of energy, entropy change conditions, and input / output boundary conditions, the generation of non-physical state estimates or control commands that may occur in general data-based artificial intelligence models can be suppressed. Accordingly, there is an effect of stably applying AI-based control correction even in power generation facilities requiring high reliability.

[0020] Furthermore, according to the aforementioned means for solving the problem of the present disclosure, the output reliability of the artificial intelligence model and the safety range of the control correction parameters can be evaluated, and if the reliability is low or an abnormal condition occurs, the correction operation can be stopped or returned to the existing control logic. Accordingly, this prevents the application of excessive control commands or unstable control corrections during the turbine control process and has the effect of ensuring the operational stability of the gas turbine.

[0021] In addition, according to the present invention, since the deterioration state or efficiency degradation state inside the turbine can be visualized through virtual sensing, there is an effect of being able to extend beyond simple control performance improvement to predictive maintenance and digital twin systems.

[0022] The effects obtainable from the exemplary embodiments of the present disclosure are not limited to those mentioned above, and other unmentioned effects can be clearly derived and understood by those skilled in the art to which the exemplary embodiments of the present disclosure belong from the description below. That is, unintended effects resulting from the implementation of the exemplary embodiments of the present disclosure can also be derived by those skilled in the art from the exemplary embodiments of the present disclosure. Brief explanation of the drawing

[0023] FIG. 1 is a block diagram showing the overall configuration of a turbine control system based on a physical law-compliant artificial neural network according to one embodiment of the present invention. FIG. 2 is a diagram showing the loss function structure of a physical law-compliant artificial neural network (PINN) model according to one embodiment of the present invention. FIG. 3 is a diagram showing a detailed configuration in which an AI correction signal is injected into the FBD control logic according to one embodiment of the present invention. FIG. 4 is a flowchart showing a parameter compensation algorithm according to an optimal output deviation according to an embodiment of the present invention. Specific details for implementing the invention

[0024] Hereinafter, various embodiments of the present disclosure are described in conjunction with the accompanying drawings. As various embodiments of the present disclosure may be subject to various modifications and may have various forms, specific embodiments are illustrated in the drawings and described in detail. However, this is not intended to limit the various embodiments of the present disclosure to specific forms, and it should be understood that they include all modifications and / or equivalents and substitutions that fall within the spirit and scope of the various embodiments of the present disclosure. In relation to the description of the drawings, similar reference numerals have been used for similar components.

[0025] In various embodiments of the present disclosure, terms such as “comprising” or “having” are intended to indicate the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0026] In various embodiments of the present disclosure, expressions such as “or” include any and all combinations of the words listed together. For example, “A or B” may include A, may include B, or may include both A and B.

[0027] Expressions such as "first," "second," "first," or "second" used in various embodiments of the present disclosure may modify various components of the various embodiments, but do not limit such components. For example, such expressions do not limit the order and / or importance of such components and may be used to distinguish one component from another.

[0028] When it is mentioned that a component is "connected" or "joined" to another component, it should be understood that the component may be directly connected or joined to the other component, but that a new component may also exist between the component and the other component.

[0029] In the embodiments of the present disclosure, terms such as "module," "unit," "part," etc. are used to refer to a component that performs at least one function or operation, and such component may be implemented in hardware or software, or in a combination of hardware and software. Additionally, a plurality of "modules," "units," "parts," etc. may be integrated into at least one module or chip and implemented as at least one processor, except where each needs to be implemented in specific individual hardware.

[0030] Terms such as those defined in commonly used dictionaries should be interpreted as having meanings consistent with their meanings in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in the various embodiments of the present disclosure.

[0031] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the attached drawings.

[0032] FIG. 1 is a block diagram showing the overall configuration of a turbine control system based on a physical law-compliant artificial neural network according to an embodiment of the present invention. Referring to FIG. 1, the turbine control system according to an embodiment of the present invention may include a turbine / generator sensor (instrument) (100), a data collection and preprocessing unit (110), a virtual sensor unit (120), an output deviation calculation unit (130), a correction parameter generation unit (140), a safety constraint and reliability evaluation unit (150), an FBD dynamic compensation module (160), and a turbine and generator (170).

[0033] The turbine / generator sensor (instrument) (100) is configured to measure the operating status of the turbine and generator (170) and can measure the pressure, temperature, flow rate, rotational speed, valve opening amount, generator output, and other operating status data of the gas turbine. The operating data measured by the turbine / generator sensor (instrument) (100) can be transmitted to the data collection and preprocessing unit (110).

[0034] The data collection and preprocessing unit (110) can collect operating data of a gas turbine from a turbine / generator sensor (instrument) (100) and preprocess the collected operating data into a form suitable for input to an artificial intelligence model. For example, the data collection and preprocessing unit (110) can perform at least one of normalization, noise removal, missing value correction, outlier correction, and time synchronization on the collected operating data. Accordingly, the data collection and preprocessing unit (110) can provide the preprocessed operating data to the virtual sensor unit (120).

[0035] The virtual sensor unit (120) can estimate thermodynamic state variables inside the turbine using a physics-compliant artificial neural network (PINN) learned based on a loss function including physics laws, with preprocessed operating data as input. Here, the thermodynamic state variables may be state variables that are difficult to measure directly or difficult to measure by a separate sensor, and may include, for example, at least one of the enthalpy of a compressor or combustor, compressor efficiency, combustor state, turbine efficiency, and entropy change amount. The virtual sensor unit (120) can transmit the estimated thermodynamic state variables to the output deviation calculation unit (130).

[0036] The output deviation calculation unit (130) calculates a theoretical output value under current operating conditions based on thermodynamic state variables received from the virtual sensor unit (120), and can calculate an output deviation by comparing the calculated theoretical output value with an actual output value obtained from a turbine / generator sensor (instrument) (100) or a generator measurement value. Here, the output deviation may be a value indicating how much output loss or efficiency reduction the current gas turbine has compared to the theoretically achievable output.

[0037] Additionally, the output deviation calculation unit (130) can analyze the time response characteristics, steady-state error, or deviation change rate of the output deviation to distinguish whether the cause of the output deviation is due to the inadequacy of the control parameters or due to the degradation of equipment performance. The output deviation calculation unit (130) can transmit the calculated output deviation or deviation analysis results to the correction parameter generation unit (140).

[0038] The correction parameter generation unit (140) can generate control correction parameters to correct the control parameters of an existing turbine controller based on the output deviation received from the output deviation calculation unit (130). The control correction parameters may include at least one of a feedforward correction value for a guide vane valve control signal, a proportional gain correction value, an integral gain correction value, a derivative gain correction value, and a control time constant correction value of a PID controller. The correction parameter generation unit (140) can transmit the generated control correction parameters to the safety constraint and reliability evaluation unit (150).

[0039] The safety constraint and reliability evaluation unit (150) can evaluate whether the control correction parameter generated by the correction parameter generation unit (140) is applicable. For example, the safety constraint and reliability evaluation unit (150) can determine whether the control correction parameter is within a pre-set safety range and determine whether to apply the control correction parameter based on the physical equation residuals calculated from the physical law-compliant artificial neural network of the virtual sensor unit (120), whether the input data is abnormal, or the reliability of the estimation result.

[0040] The safety constraint and reliability evaluation unit (150) can transmit the verified control correction parameter to the FBD dynamic compensation module (160) when the control correction parameter is within a safe range and the output reliability of the physical law-compliant artificial neural network is greater than or equal to a preset threshold. Conversely, when the control correction parameter is outside the safe range or the output reliability of the physical law-compliant artificial neural network is below the threshold, the safety constraint and reliability evaluation unit (150) can stop the application of the control correction parameter or control it to return to the existing control logic.

[0041] The FBD dynamic compensation module (160) can control the turbine and generator (170) by reflecting control correction parameters verified by the safety constraint and reliability evaluation unit (150) into the functional block diagram (FBD)-based control logic. For example, the FBD dynamic compensation module (160) can add a feedforward correction value to the valve control command value generated from the existing FBD control logic, or dynamically change the proportional gain, integral gain, or derivative gain of the PID controller. Accordingly, the FBD dynamic compensation module (160) can compensate and control the operating state of the turbine and generator (170) in real time by reflecting the state information and output deviation calculated by the virtual sensor unit (120) and the output deviation calculation unit (130), while maintaining the existing FBD-based control structure. Therefore, the turbine control system of the present invention can be configured to inject additional correction signals without completely replacing the existing controller, thereby improving control performance while ensuring applicability to existing power generation facilities.

[0042] The turbine and generator (170) can be operated according to control commands generated or corrected by the FBD dynamic compensation module (160). The operation results of the turbine and generator (170) are measured again by the turbine / generator sensor (instrument) (100), and the measured operation data can be transmitted back to the data collection and preprocessing unit (110). In this way, the turbine control system of the present invention can form a closed-loop control structure that utilizes the operation results of the turbine and generator (170) as input data.

[0043] Accordingly, a turbine control system according to one embodiment of the present invention can improve the operating efficiency and control stability of a gas turbine by using a physical law-compliant artificial neural network to estimate thermodynamic state variables inside the turbine that are difficult to measure directly, calculating an output deviation based on the estimated thermodynamic state variables, generating control correction parameters according to the calculated output deviation, and reflecting the generated control correction parameters in the existing FBD-based control logic after evaluating safety and reliability.

[0044] The data collection and preprocessing unit (110) collects at least one of the gas pressure, gas temperature, gas flow rate, and generator output data of the gas turbine in real time from a turbine / generator sensor (instrument) (100), and performs at least one of normalization, noise removal, and outlier correction on the collected operation data. In an embodiment, the data collection and preprocessing unit (110) can collect operation data of the gas turbine in real time from a turbine / generator sensor (instrument) (100). Here, the operation data may include at least one of the gas pressure, gas temperature, gas flow rate, and generator output data of the gas turbine.

[0045] The turbine / generator sensor (instrument) (100) may include a plurality of sensors for measuring the operating status of a gas turbine or generator, for example, a pressure sensor, a temperature sensor, a flow sensor, an output meter, and other operating status meters. The data collection and preprocessing unit (110) receives operating data collected in chronological order from the turbine / generator sensor (instrument) (100) and can convert the received operating data into an input data form that can be processed by the virtual sensor unit (120).

[0046] Additionally, the data collection and preprocessing unit (110) can perform at least one of normalization, noise removal, and outlier correction on the collected operation data. For example, the data collection and preprocessing unit (110) can normalize gas pressure, gas temperature, gas flow rate, and generator output data having different units or ranges into a preset reference range. Accordingly, learning or estimation errors caused by scale differences in data input to a physical law-compliant artificial neural network (PINN) can be reduced.

[0047] Additionally, the data collection and preprocessing unit (110) may perform filtering processing to remove instantaneous noise or measurement errors included in the sensor signal. For example, the data collection and preprocessing unit (110) may remove or mitigate noise components that fluctuate rapidly on the time axis, thereby enabling the virtual sensor unit (120) to receive stable operating state information.

[0048] Additionally, the data collection and preprocessing unit (110) determines whether there are any outliers among the collected operating data that fall outside the normal operating range, and if outliers are detected, it can correct or exclude the corresponding data. For example, the data collection and preprocessing unit (110) can determine outliers based on correlations with other operating variables in the same time period, changes in operating data in previous time intervals, or pre-set allowable ranges. In this way, the data collection and preprocessing unit (110) can improve the accuracy of estimating internal thermodynamic state variables of the turbine using a physical law-compliant artificial neural network by refining the operating data collected in real time from the turbine / generator sensor (instrument) (100) and providing it to the virtual sensor unit (120).

[0049] In the embodiment, the physical law-compliant artificial neural network is included in the virtual sensor unit (120) and can be trained based on a loss function comprising at least one of a data loss term corresponding to the error between the operating data and the predicted value, a physical loss term reflecting the energy conservation law of the gas turbine, a constraint term reflecting the entropy change condition, and a boundary loss term reflecting the input / output boundary condition. In the embodiment, the virtual sensor unit (120) may include a physical law-compliant artificial neural network (PINN). The physical law-compliant artificial neural network can be trained to estimate thermodynamic state variables inside the gas turbine using preprocessed operating data provided from the data collection and preprocessing unit (110) as input.

[0050] The above-described law-compliant artificial neural network can be trained not merely by learning the statistical correlation between operating data and output values, but by incorporating the physical laws that must be satisfied during the operation of the gas turbine into the loss function. Accordingly, the law-compliant artificial neural network can estimate state variables within a physically acceptable range even under operating conditions that fall outside the range of the training data. Specifically, the loss function of the law-compliant artificial neural network may include at least one of a data loss term, a physical loss term, a constraint term, and a boundary loss term. The data loss term may represent the error between the actual measured value or reference value corresponding to the operating data and the predicted value of the law-compliant artificial neural network. Through this, the law-compliant artificial neural network can be trained in a direction that aligns with the actual gas turbine operating data.

[0051] The above physical loss term may be a term that reflects the law of conservation of energy of a gas turbine. For example, the above physical loss term may be configured so that the relationship of energy change of the fluid passing through the compressor, combustor, and turbine satisfies a pre-established thermodynamic relationship. Accordingly, the above physical law-compliant artificial neural network can suppress the generation of non-physical state estimates that violate the law of conservation of energy.

[0052] The above constraint term may be a term reflecting the entropy change condition. For example, the constraint term may be set based on the range of entropy change allowed in the thermodynamic process inside the gas turbine, and may be configured to increase the loss value if the physics-law-compliant artificial neural network outputs an estimate that does not meet the entropy change condition. Accordingly, the physics-law-compliant artificial neural network can provide thermodynamically valid state variable estimation results.

[0053] The above boundary loss term may be a term that reflects the input / output boundary conditions of the gas turbine. For example, the above boundary loss term may be set based on boundary conditions regarding at least one of the gas turbine's inlet pressure, inlet temperature, outlet pressure, outlet temperature, flow rate, and generator output. Accordingly, the above physical law-compliant artificial neural network can prevent the generation of state estimates that deviate from the input / output conditions permitted in the actual gas turbine operating environment. In this way, the physical law-compliant artificial neural network included in the virtual sensor unit (120) is trained based on a loss function that reflects not only the data loss term but also the law of conservation of energy, entropy change conditions, and input / output boundary conditions, thereby enabling the estimation of thermodynamic state variables inside the turbine, which are difficult to measure directly, while maintaining physical consistency.

[0054] Additionally, the virtual sensor unit (120) estimates at least one of the enthalpy, compressor efficiency, combustor state, turbine efficiency, and entropy change amount of the compressor or combustor, which are difficult to measure directly, as the thermodynamic state variable using the physical law-compliant artificial neural network. In an embodiment, the virtual sensor unit (120) can estimate the thermodynamic state variable inside the gas turbine using the physical law-compliant artificial neural network (PINN). Here, the thermodynamic state variable may refer to internal state information that is difficult to measure directly during the operation of the gas turbine, or which is difficult to measure accurately even if a separate sensor is installed.

[0055] For example, the virtual sensor unit (120) receives at least one of gas pressure, gas temperature, gas flow rate, and generator output data provided by the data collection and preprocessing unit (110) as input, and applies the input data to a physical law-compliant artificial neural network to estimate at least one of the enthalpy, compressor efficiency, combustor state, turbine efficiency, and entropy change amount of the compressor or combustor as a thermodynamic state variable.

[0056] The enthalpy of the compressor or combustor may be a variable representing the energy state of the fluid inside the gas turbine. The virtual sensor unit (120) can estimate the enthalpy state before and after the compressor or combustor based on the relationship of changes in pressure, temperature, and flow rate, thereby identifying the state of energy change during the compression or combustion process. Compressor efficiency may be a variable representing the relationship between the compression work theoretically required during the process of the compressor compressing incoming air and the compression performance in the actual operating state. The virtual sensor unit (120) can estimate the compressor efficiency based on changes in pressure and temperature at the compressor inlet and outlet, changes in flow rate, or changes in generator output. Accordingly, changes in efficiency due to compressor performance degradation or changes in operating conditions can be identified in real time.

[0057] The combustor state may include internal states related to the combustor's combustion efficiency, heat generation state, combustion stability, or fuel supply state. The virtual sensor unit (120) can estimate the combustor state based on temperature changes, pressure changes, and flow rate changes before and after the combustor, thereby providing information necessary to determine whether combustion instability, efficiency degradation, or thermal loss has occurred. Turbine efficiency may be a variable representing the efficiency with which a turbine converts the energy of high-temperature and high-pressure gas into mechanical rotational energy. The virtual sensor unit (120) can estimate turbine efficiency based on the thermodynamic state of the turbine inlet and outlet and generator output data, and the estimated turbine efficiency can be used to calculate theoretical output values ​​under current operating conditions or to analyze output deviations.

[0058] The change in entropy may be a variable indicating the irreversibility or degree of loss of the internal thermodynamic process of the gas turbine. The virtual sensor unit (120) can estimate the change in entropy in the compressor, combustor, or turbine section based on the relationship between pressure, temperature, and flow rate. Accordingly, thermodynamic losses or abnormal operating conditions occurring inside the gas turbine can be identified more precisely. In this way, the virtual sensor unit (120) can estimate the internal state of the gas turbine, which is difficult to verify with actual sensors alone, by virtually sensing internal thermodynamic state variables that are difficult to measure directly through a physical law-compliant artificial neural network. Additionally, the thermodynamic state variables estimated by the virtual sensor unit (120) are provided to the output deviation calculation unit (130) and can be used to calculate the theoretical output value under current operating conditions and to analyze the deviation from the actual output value.

[0059] Additionally, the output deviation calculation unit (130) analyzes at least one of the time response characteristics and steady-state error of the output deviation and classifies the cause of the output deviation as either inappropriateness of control parameters or degradation of equipment performance based on the analysis results. In an embodiment, the output deviation calculation unit (130) can calculate a theoretical output value under current operating conditions based on thermodynamic state variables estimated by the virtual sensor unit (120) and calculate an output deviation between the theoretical output value and the actual output value. Here, the actual output value may be a generator output value obtained from a turbine / generator sensor (instrument) (100) or a generator output meter.

[0060] The output deviation calculation unit (130) may analyze at least one of the time response characteristics and steady-state error of the output deviation to determine whether the output deviation is due to simple temporary fluctuations, inappropriateness of control parameters, or equipment performance degradation. For example, the output deviation calculation unit (130) may analyze whether the output deviation decreases over time, persists for a certain period of time, or vibrates repeatedly after a change in the load command, fuel supply amount, or guide vane valve opening amount. In this case, at least one of the rise time, settling time, overshoot, undershoot, degree of damping, and deviation change rate of the output deviation may be used as the time response characteristics.

[0061] Additionally, the output deviation calculation unit (130) can calculate the error remaining between the theoretical output value and the actual output value as a steady-state error even after the operating state of the gas turbine has reached a steady state. If the steady-state error exceeds a preset allowable range, the output deviation calculation unit (130) can determine that the current control parameter does not sufficiently reflect the actual operating characteristics of the turbine.

[0062] In one embodiment, the output deviation calculation unit (130) may classify the cause of the output deviation as the inadequacy of control parameters, such as PID gain, control time constant, or feedforward correction value, when the output deviation occurs significantly immediately after a change in control input and then gradually decreases, or occurs repeatedly in a specific control interval. The result classified in this way may be transmitted to the correction parameter generation unit (140) and used to generate control correction parameters.

[0063] In another embodiment, the output deviation calculation unit (130) may classify the cause of the output deviation as equipment performance degradation if the output deviation continues to increase despite the correction of control parameters, if the steady-state error is maintained for a long period, or if the actual output value repeatedly decreases under the same operating conditions. Here, equipment performance degradation may include compressor performance degradation, combustor efficiency degradation, turbine blade deterioration, increased leakage, or other mechanical deterioration conditions.

[0064] Additionally, the output deviation calculation unit (130) can classify the cause of the output deviation by considering the time response characteristics and steady-state error together. For example, if the output deviation occurs only in the transient state and is resolved in the steady state, it can be determined as a temporary control response delay, and if the output deviation is maintained even in the steady state, it can be determined as inappropriateness of the control parameters or degradation of equipment performance. In this way, the output deviation calculation unit (130) does not stop at calculating the output deviation between the theoretical output value and the actual output value, but can classify the cause of the output deviation as either inappropriateness of the control parameters or degradation of equipment performance by analyzing the time response characteristics and steady-state error of the output deviation. Accordingly, the correction parameter generation unit (140) can generate a control correction parameter suitable for the cause of the output deviation, and the turbine control system can perform more appropriate compensation control according to the cause of the reduction in operating efficiency.

[0065] Additionally, the correction parameter generation unit (140) generates at least one of a feedforward correction value for a guide vane valve control signal, a proportional gain correction value, an integral gain correction value, and a derivative gain correction value for a PID controller as the control correction parameter based on the output deviation. In an embodiment, the correction parameter generation unit (140) can generate a control correction parameter to correct the control parameter of an existing turbine controller based on the output deviation or the result of classifying the cause of the output deviation transmitted from the output deviation calculation unit (130). Here, the output deviation may refer to the difference between the theoretical output value and the actual output value calculated under current operating conditions, and the correction parameter generation unit (140) can calculate the control correction parameter by considering at least one of the magnitude, sign, rate of change, and duration of the output deviation.

[0066] In one embodiment, the correction parameter generation unit (140) can generate a feedforward correction value for a guide vane valve control signal when it is determined that the output deviation is related to the opening control of the guide vane valve. For example, if the actual output value is lower than the theoretical output value, the correction parameter generation unit (140) can generate a feedforward correction value to correct the target opening amount of the guide vane valve or the valve control command value. The feedforward correction value can be used as a correction signal that is added to or subtracted from the valve control command value calculated in the existing FBD-based control logic.

[0067] Additionally, the correction parameter generation unit (140) can generate at least one of a proportional gain correction value, an integral gain correction value, and a derivative gain correction value of the PID controller based on the time response characteristics of the output deviation. For example, if the output deviation does not decrease sufficiently immediately in response to a change in the control input, the correction parameter generation unit (140) can generate a proportional gain correction value to adjust the control responsiveness. Additionally, if a steady-state error persists, the correction parameter generation unit (140) can generate an integral gain correction value to reduce the residual deviation. Additionally, if the output deviation fluctuates rapidly or contains vibration components, the correction parameter generation unit (140) can generate a derivative gain correction value to improve the stability of the output response.

[0068] Additionally, the correction parameter generation unit (140) can generate control correction parameters to prioritize gain correction or feedforward correction of the PID controller when the cause of the output deviation is classified as inappropriateness of the control parameters. On the other hand, when the cause of the output deviation is classified as equipment performance degradation, the correction parameter generation unit (140) can generate correction parameters within a limited range so that excessive control compensation is not applied, or provide relevant information so that the safety constraint and reliability evaluation unit (150) can determine whether to apply the correction.

[0069] In one embodiment, the control correction parameter may be any one of a feedforward correction value for a guide vane valve control signal, a proportional gain correction value, an integral gain correction value, and a derivative gain correction value of a PID controller, or a combination of two or more of these. Additionally, the control correction parameter may not be a value that directly replaces the existing control parameter, but may be generated in the form of a correction coefficient that is added to, subtracted from, or multiplied by the existing control parameter. In this way, the correction parameter generation unit (140) generates a feedforward correction value for a guide vane valve control signal or a gain correction value of a PID controller based on the output deviation, thereby enabling dynamic control compensation corresponding to changes in the operating state of the gas turbine while maintaining the structure of the existing FBD-based control logic.

[0070] The above safety constraint and reliability evaluation unit (150) determines whether the control correction parameter is within a preset safety range and evaluates the output reliability of the physics-law-compliant artificial neural network based on whether there is an abnormality in the residual of the physics equation or the input data of the physics-law-compliant artificial neural network. If the control correction parameter is outside the safety range or the output reliability is below a preset threshold, the application of the control correction parameter is stopped and the existing FBD-based control logic is returned. In an embodiment, the safety constraint and reliability evaluation unit (150) can evaluate whether the control correction parameter generated by the correction parameter generation unit (140) is applicable to actual gas turbine control. Before the control correction parameter is reflected in the existing FBD-based control logic, the safety constraint and reliability evaluation unit (150) can determine whether at least one of the magnitude, rate of change, application direction, and application duration of the control correction parameter is within a preset safety range.

[0071] The above safety range may be set based on at least one of the operating stability of the gas turbine, the allowable opening range of the guide vane valve, the allowable gain range of the PID controller, the allowable thermal stress range of the turbine, the allowable range of generator output fluctuation, and the facility protection conditions. For example, if the control correction parameter is a feedforward correction value for the guide vane valve control signal, the safety constraint and reliability evaluation unit (150) can determine whether the valve command value after correction exceeds the allowable opening range. Additionally, if the control correction parameter is a gain correction value of the PID controller, the safety constraint and reliability evaluation unit (150) can determine whether the proportional gain, integral gain, or derivative gain after correction is within a preset allowable gain range.

[0072] Additionally, the safety constraint and reliability evaluation unit (150) can evaluate the output reliability of the physical law-compliant artificial neural network included in the virtual sensor unit (120). In one embodiment, the safety constraint and reliability evaluation unit (150) can calculate the output reliability based on the physical equation residuals of the physical law-compliant artificial neural network. Here, the physical equation residuals may be values ​​indicating the extent to which the thermodynamic state variables estimated by the physical law-compliant artificial neural network satisfy the energy conservation law, entropy change conditions, or input / output boundary conditions.

[0073] For example, if the physical equation residual is less than or equal to a preset reference value, the safety constraint and reliability evaluation unit (150) may determine that the estimation result of the virtual sensor unit (120) conforms to the physical law. Conversely, if the physical equation residual exceeds a preset reference value, the safety constraint and reliability evaluation unit (150) may determine that the estimation result of the virtual sensor unit (120) is physically unreliable and may limit the application of the control correction parameter generated based on the estimation result.

[0074] Additionally, the safety constraint and reliability evaluation unit (150) can evaluate the output reliability of the physical law-compliant artificial neural network based on whether there are abnormalities in the input data. For example, if the gas pressure, gas temperature, gas flow rate, or generator output data provided by the data collection and preprocessing unit (110) is outside the normal operating range, or if it contains sensor defects, communication errors, missing values, or sudden abnormal fluctuations, the safety constraint and reliability evaluation unit (150) can evaluate the output reliability of the virtual sensor unit (120) based on the input data as low. If the control correction parameter is within a preset safety range and the output reliability of the physical law-compliant artificial neural network exceeds a preset threshold, the safety constraint and reliability evaluation unit (150) can determine the control correction parameter as a verified correction parameter and transmit it to the FBD dynamic compensation module (160). Accordingly, the FBD dynamic compensation module (160) can compensate and control the gas turbine by reflecting the verified control correction parameter in the existing FBD-based control logic.

[0075] On the other hand, the safety constraint and reliability evaluation unit (150) may stop applying the control correction parameter when the control correction parameter exceeds a preset safety range or when the output reliability of the physical law-compliant artificial neural network is below a preset threshold. In this case, the safety constraint and reliability evaluation unit (150) may transmit a correction stop signal or a return control signal to the FBD dynamic compensation module (160), and the FBD dynamic compensation module (160) may control the gas turbine according to the existing FBD-based control logic without reflecting the control correction parameter.

[0076] In one embodiment, the return to the existing FBD-based control logic can be performed using a shock-free switching method. That is, the FBD dynamic compensation module (160) can immediately block the application of the control correction parameter, or gradually reduce the correction value for a certain period of time and then return to the existing control logic. Accordingly, it is possible to prevent the valve command value, generator output, or turbine rotation state from fluctuating rapidly during the process of stopping the application of the control correction parameter. In this way, the safety constraint and reliability evaluation unit (150) evaluates the safety range of the control correction parameter and the output reliability of the physical law-compliant artificial neural network together, thereby preventing correction values ​​that are physically unreliable or may exceed the facility operating limits from being applied to the actual turbine control. Therefore, the turbine control system of the present invention can secure the operational stability and reliability required for power generation facilities while performing artificial intelligence-based correction control.

[0077] FIG. 2 is a diagram showing the structure of a loss function of a physical law-compliant artificial neural network (PINN) model according to an embodiment of the present invention. Referring to FIG. 2, a physical law-compliant artificial neural network included in a virtual sensor unit (120) can be trained using input data (210), a PINN model (220), a prediction output (230), physical law constraints (240), physical law loss (250), data loss (260), a total loss function (270), and an optimization unit (280).

[0078] The input data (210) may include real-time measured values ​​and operating condition parameters representing the operating status of the gas turbine. For example, the real-time measured values ​​may include at least one of fuel flow rate, compressor inlet temperature, compressor inlet pressure, ambient temperature, ambient pressure, rotational speed, and guide vane opening. Additionally, the operating condition parameters may include load commands, environmental conditions, and other operating condition information. These input data (210) may be input into the PINN model (220) after being collected and preprocessed by the data collection and preprocessing unit (110).

[0079] The PINN model (220) may be an artificial neural network model that estimates physical quantities related to the thermodynamic state or performance inside a gas turbine using input data (210). Unlike a general data-based artificial neural network, the PINN model (220) may be trained by considering the thermodynamic laws of the gas turbine, performance maps, and physical law constraints based on the equation of state. Accordingly, the PINN model (220) may be trained not only to reduce the prediction error on the training data but also to generate a prediction output (230) within a physically valid range.

[0080] The predicted output (230) may refer to physical quantities calculated by the PINN model (220). For example, the predicted output (230) may include at least one of the compressor outlet temperature, compressor outlet pressure, combustor outlet temperature, turbine inlet temperature, turbine outlet pressure, shaft work, fuel quantity, and exhaust gas temperature. This predicted output (230) may be used to estimate thermodynamic state variables inside a gas turbine that are difficult to measure directly or difficult to install separate sensors for.

[0081] The physical law constraint (240) may include constraints to ensure that the predicted output (230) of the PINN model (220) corresponds to the physical behavior of the gas turbine. In one embodiment, the physical law constraint (240) may include at least one of a mass conservation constraint, an energy conservation constraint, a compressor performance map constraint, a turbine performance map constraint, a combustor energy balance constraint, and an equation of state constraint.

[0082] Mass conservation constraints can be set so that the relationship between the mass flow rate of the working fluid entering the gas turbine and the mass flow rate of the working fluid exiting the gas turbine satisfies the continuity equation. Accordingly, the PINN model (220) can be prevented from generating non-physical flow rate predictions that violate the law of mass conservation.

[0083] Energy conservation constraints can be set so that the relationship between the thermal energy, work, and enthalpy change of the working fluid passing through the compressor, combustor, and turbine satisfies the first law of thermodynamics. Accordingly, the PINN model (220) can generate a prediction result that corresponds to the energy input / output relationship inside the gas turbine.

[0084] The compressor performance map constraint can be set so that the relationship between the compressor's pressure ratio, rotational speed, and efficiency matches the preset compressor performance map. The turbine performance map constraint can be set so that the relationship between the turbine inlet temperature, pressure ratio, rotational speed, and turbine efficiency matches the preset turbine performance map. Accordingly, the PINN model (220) can predict state variables that reflect the performance characteristics of the actual gas turbine device.

[0085] The combustor energy balance constraint can be set to satisfy the energy balance relationship between the lower heating value of the fuel, the air flow rate, the fuel flow rate, and the combustor outlet enthalpy. Accordingly, the PINN model (220) can physically and validly estimate the heat generation and energy transfer state inside the combustor.

[0086] The equation of state constraints can be set so that the relationship between pressure, density, gas constant, and temperature satisfies the ideal gas equation of state. Accordingly, the PINN model (220) can be trained so that the predicted pressure and temperature values ​​inside the gas turbine correspond to the thermodynamic equation of state.

[0087] The physical law loss (250) can be calculated by weighting a plurality of loss terms corresponding to the physical law constraint (240). For example, the physical law loss (250) may include at least one of mass conservation loss, energy conservation loss, compressor performance map loss, turbine performance map loss, combustor energy balance loss, and equation of state loss. Each loss term may be assigned a weight according to the importance of the physical constraint.

[0088] Data loss (260) may represent an error between the predicted output (230) of the PINN model (220) and the actual measurable output value. For example, data loss (260) may be calculated based on the mean squared error between the actual measurable output value, such as exhaust gas temperature, shaft work, and pressure, and the predicted value of the PINN model (220). Accordingly, the PINN model (220) can be trained to match the actual driving data.

[0089] The total loss function (270) can be calculated by considering both the data loss (260) and the physical law loss (250). For example, the total loss function (270) can be configured in the form of adding the physical law loss to the data loss with weights, and the weights for the physical law loss can be set as tuning parameters. Accordingly, the PINN model (220) can be trained to simultaneously satisfy prediction accuracy for actual data and consistency with physical laws. The optimization unit (280) can update the learning parameters of the PINN model (220) so that the total loss function (270) is reduced. In one embodiment, the optimization unit (280) may use an Adam optimizer or a similar gradient-based optimization algorithm. The optimization unit (280) can train a virtual sensor model that predicts the internal state of a gas turbine by updating the weights and biases of the PINN model (220) in a direction that minimizes the total loss function (270).

[0090] Additionally, as shown in the description section (290) of FIG. 2, in the loss function structure, N represents the number of samples, M represents the number of predicted output dimensions, λ represents the weight of each physical constraint term, and β represents the physical loss weight. These weights can be adjusted according to the operating characteristics of the gas turbine, the quality of the training data, the importance of the physical constraints, and the purpose of application of the control system. Thus, the PINN model (220) according to one embodiment of the present invention is trained based on a total loss function (270) that reflects both the data loss (260) and the physical law loss (250), thereby providing physically valid prediction results compared to a simple data-based model. Accordingly, the virtual sensor unit (120) can more reliably estimate thermodynamic state variables inside the gas turbine that are difficult to measure directly, and the estimated state variables can be provided to the output deviation calculation unit (130) and the correction parameter generation unit (140) to be utilized for dynamic compensation of turbine control parameters.

[0091] FIG. 3 is a diagram showing a detailed configuration in which an AI correction signal is injected into the FBD control logic according to one embodiment of the present invention.

[0092] Referring to FIG. 3, a turbine control system according to one embodiment of the present invention may include a measurement and input signal unit (310), an FBD-based gas turbine control logic (320), an AI correction module (330), an AI correction signal injection unit (350), a drive / actuator (360), and a gas turbine (340). FIG. 3 illustrates a configuration in which an AI correction signal (334) generated by a PINN-based AI correction module (330) is additionally injected into an existing control command while maintaining an existing DCS or PLC-based FBD control structure.

[0093] The measurement and input signal unit (310) can collect key measurement values ​​of the gas turbine (340). For example, the measurement and input signal unit (310) can measure at least one of the combustor inlet temperature, combustor pressure, combustor outlet temperature, turbine inlet temperature, turbine outlet temperature, exhaust pressure, rotational speed, fuel flow rate, and IGV angle. Such actual measurement data can be provided to the FBD-based gas turbine control logic (320) and simultaneously provided as input data to the AI ​​correction module (330).

[0094] The FBD-based gas turbine control logic (320) may be an existing turbine control logic implemented in a DCS or PLC. The FBD-based gas turbine control logic (320) receives at least one of a load command, a rotational speed command, and a combustor temperature command, and based thereon, can generate at least one of a fuel flow rate command (324), an IGV angle command (325), and a combustor fuel distribution or fuel schedule command (326).

[0095] In one embodiment, the FBD-based gas turbine control logic (320) may include a load controller (321), a rotational speed controller (322), and a combustor temperature controller (323). The load controller (321) may generate a fuel flow rate command (324) based on a load command, the rotational speed controller (322) may generate an IGV angle command (325) based on a rotational speed command, and the combustor temperature controller (323) may generate a combustor fuel distribution or fuel schedule command (326) based on a combustor temperature command. At this time, the load controller (321), the rotational speed controller (322), and the combustor temperature controller (323) may each include a PID control block.

[0096] Control commands generated in the FBD-based gas turbine control logic (320) can be transmitted to a drive / actuator (360). For example, a fuel flow command (324) can be transmitted to a fuel control valve (361), an IGV angle command (325) can be transmitted to an IGV actuator (362), and a combustor fuel distribution or fuel schedule command (326) can be transmitted to a combustor fuel valve (363). Accordingly, the gas turbine (340) can be operated according to the control commands.

[0097] The AI ​​correction module (330) may be a PINN-based correction module. The AI ​​correction module (330) receives actual measurement data provided from the measurement and input signal unit (310) and can generate an AI correction signal (334) using the virtual sensing and state estimation unit (331), the deviation analysis and cause diagnosis unit (332), and the correction signal generation unit (333).

[0098] In one embodiment, the virtual sensing and state estimation unit (331) can estimate the thermodynamic state inside the gas turbine (340) using a PINN model. For example, the virtual sensing and state estimation unit (331) can estimate at least one of combustor efficiency, turbine efficiency, combustion efficiency, flow rate characteristics, and internal heat transfer characteristic values. Here, the internal heat transfer characteristic value may include at least one of temperature, pressure, pressure ratio, and enthalpy-related state values.

[0099] The deviation analysis and cause diagnosis unit (332) can calculate a theoretical model-based output prediction value based on the state information estimated by the virtual sensing and state estimation unit (331) and calculate an output deviation by comparing it with the actual output. Additionally, the deviation analysis and cause diagnosis unit (332) can analyze whether the cause of the deviation is due to any one of the control offset, control delay, gain impropriety, or equipment deterioration based on the calculated output deviation. Accordingly, the AI ​​correction module (330) can calculate the control parameters and correction amounts that require correction.

[0100] The correction signal generation unit (333) can generate an AI correction signal (334) based on the analysis results of the deviation analysis and cause diagnosis unit (332). The AI ​​correction signal (334) may include at least one of a fuel flow rate correction value, an IGV angle correction value, a fuel schedule correction value, and other correction parameters. For example, the correction signal generation unit (333) may generate at least one of ΔWf for correcting the fuel flow rate command (324), ΔIGV for correcting the IGV angle command (325), and ΔFuelSch for correcting the combustor fuel distribution or fuel schedule command (326).

[0101] The AI ​​correction signal injection unit (350) can inject the AI ​​correction signal (334) generated by the AI ​​correction module (330) into the output command of the existing FBD-based gas turbine control logic (320). For example, the AI ​​correction signal injection unit (350) can add ΔWf to the fuel flow rate command (324) generated by the existing FBD control logic, add ΔIGV to the IGV angle command (325), or add ΔFuelSch to the combustor fuel distribution or fuel schedule command (326). Accordingly, the existing control command can be converted into a modified control command that reflects the AI ​​correction signal (334).

[0102] The injection of such AI correction signals can be performed using a feedforward or bias correction method. That is, the AI ​​correction signal injection unit (350) can operate by adding a correction value to the control command calculated from the existing FBD-based gas turbine control logic (320) without replacing the existing FBD-based gas turbine control logic (320) itself. Therefore, dynamic compensation according to the operating state of the gas turbine (340) is possible while minimizing changes to the existing DCS or PLC-based control structure.

[0103] The gas turbine (340) can be operated according to a control command in which an AI correction signal (334) is injected. The actual output, rotational speed, turbine inlet temperature, exhaust gas temperature, and exhaust pressure of the gas turbine (340) can be measured again as a feedback signal (370) and transmitted to the FBD-based gas turbine control logic (320) and the AI ​​correction module (330). Accordingly, the turbine control system of the present invention can form a closed-loop control structure that repeatedly updates the existing FBD control and AI correction control by reflecting the actual operation results.

[0104] As illustrated in FIG. 3, the entire operation flow (380) can be performed in the order of collecting measurement data, generating a control output based on FBD control logic, estimating the state and analyzing deviations by an AI model, generating a correction signal, injecting the correction signal into the FBD output, applying the corrected command to the actuator, and providing actual output feedback and retuning improvement.

[0105] Additionally, according to the configuration of FIG. 3, a control signal path, a feedback signal path, and an AI correction signal path may be formed separately. The control signal path may be a path in which a control command generated from the FBD-based gas turbine control logic (320) is transmitted to the gas turbine (340) through the drive / actuator (360). The feedback signal path may be a path in which the actual operating result of the gas turbine (340) is measured as a feedback signal (370) and returned to the FBD-based gas turbine control logic (320) or the AI ​​correction module (330). The AI ​​correction signal path may be a path in which an AI correction signal (334) generated from the AI ​​correction module (330) is transmitted to the AI ​​correction signal injection unit (350) and reflected in the existing control command.

[0106] As illustrated in the feature section (390) of FIG. 3, the configuration according to one embodiment of the present invention can inject an AI correction signal (334) generated by a PINN-based AI correction module (330) in a feedforward or bias manner while minimizing changes to the existing FBD structure. Accordingly, it is possible to compensate for the degradation of control performance due to the aging of the gas turbine (340), changes in ambient conditions, or load fluctuations, and to improve the operating efficiency and control reliability of the gas turbine while maintaining the stability of the existing control system.

[0107] Below, we will look at Fig. 4. The turbine control method based on a physical law-compliant artificial neural network illustrated in Fig. 4 can be performed by a turbine control system based on a physical law-compliant artificial neural network.

[0108] Meanwhile, FIG. 4 is merely illustrative, and the concept of the present invention is not to be interpreted as being limited to that shown in FIG. 4. For example, each step may be configured in a different order than that shown in FIG. 4, at least one of the steps shown in FIG. 4 may not be performed, or one or more steps not shown in FIG. 4 may be additionally performed.

[0109] FIG. 4 is a flowchart illustrating a parameter compensation algorithm based on optimal output deviation according to an embodiment of the present invention. Referring to FIG. 4, the parameter compensation algorithm according to an embodiment of the present invention can be performed by collecting real-time operation data of a gas turbine, estimating a theoretical output value in the current operating state using a virtual sensor model, and then dynamically compensating control parameters based on the deviation from the actual output value.

[0110] First, a data collection and state estimation step (S110) may be performed. In the data collection and state estimation step (S110), the data collection and preprocessing unit may collect real-time operating data from the gas turbine, such as pressure, temperature, rotational speed, fuel flow rate, RPM, and exhaust gas composition. The collected operating data may be input into a PINN-based virtual sensor model or a PHM-based virtual sensor model, and the virtual sensor model may estimate the internal state of the gas turbine and the theoretical output value under current operating conditions based on the input data.

[0111] In one embodiment, the theoretical output value estimated in the data collection and state estimation step (S110) may be calculated based on state variables of the gas turbine, such as temperature, pressure, rotational speed, flow rate, composition information, and thermodynamic state information. Additionally, the actual output value of the gas turbine may be obtained from the generator output measurement value or the turbine output measurement value.

[0112] Next, an actual output measurement step (S120) may be performed. In the actual output measurement step (S120), the output deviation calculation unit may obtain the actual output value of the current gas turbine from the generator output measurement value or the turbine output measurement value. The actual output value may be used as a reference value for comparison with the theoretical output value calculated in the data collection and state estimation step (S110).

[0113] Next, an output deviation calculation step (S130) may be performed. In the output deviation calculation step (S130), the output deviation calculation unit may calculate the output deviation by comparing the actual output value with the theoretical output value. For example, the output deviation may be calculated as the value obtained by subtracting the theoretical output value from the actual output value.

[0114] Next, a normalized deviation calculation step (S140) may be performed. In the normalized deviation calculation step (S140), the output deviation calculation unit can calculate the normalized deviation by normalizing the calculated output deviation to a theoretical output value. Accordingly, the relative magnitude of the output deviation can be compared even under different output ranges or load conditions.

[0115] Next, a deviation tolerance range determination step (S150) may be performed. In the deviation tolerance range determination step (S150), the output deviation calculation unit may determine whether the calculated normalized deviation is within a preset tolerance range. If the normalized deviation is within the tolerance range, the current control parameter or correction parameter may be maintained, and the process may proceed to a performance verification step or the next control cycle. Conversely, if the normalized deviation exceeds the tolerance range, a deviation cause analysis step (S160) may be performed to analyze the cause of the output deviation.

[0116] In the deviation cause analysis step (S160), sensitivity-based analysis may be performed. Specifically, the output deviation calculation unit or the correction parameter generation unit may calculate sensitivity, which indicates how sensitively the theoretical output value changes with respect to changes in each control parameter. For example, sensitivity may be calculated as the partial derivative of the theoretical output value with respect to a specific parameter.

[0117] Additionally, in the deviation cause analysis step (S160), at least one of turbine efficiency, GV angle, compressor guide vane angle, combustor fuel distribution ratio, turbine inlet temperature, compressor pressure ratio, rotational speed, combustor temperature scale factor, and cooling air flow rate may be set as a parameter to be analyzed. Among the parameters to be analyzed, the top parameter with a significant influence on the output deviation may be identified as the primary parameter to be compensated.

[0118] Next, an optimal compensation amount calculation step (S170) may be performed. In the optimal compensation amount calculation step (S170), the correction parameter generation unit may calculate an optimal compensation amount with the goal of minimizing the output deviation. In one embodiment, the correction parameter generation unit may set an objective function that minimizes the output deviation or normalization deviation while ensuring that the amount of change in the correction parameter does not become excessively large.

[0119] Additionally, in the optimal compensation amount calculation step (S170), the correction parameter generation unit may reflect at least one of the lower and upper limits of each control parameter, a turbine inlet temperature limit, an exhaust gas temperature limit, a pressure ratio limit, and mechanical or thermal stability conditions as a constraint. Accordingly, the correction parameter generation unit may calculate the optimal compensation amount using gradient descent, Newton's method, quadratic programming, or a similar optimization technique.

[0120] Next, a step for verifying the applicability of the optimal compensation amount (S180) may be performed. In the step for verifying the applicability of the optimal compensation amount (S180), the correction parameter generation unit or the safety constraint and reliability evaluation unit may determine whether the calculated optimal compensation amount satisfies the constraints and physical allowable range. If it is determined that the calculated optimal compensation amount is applicable, a parameter compensation application step (S190) may be performed.

[0121] In one embodiment, the optimal compensation amount may be calculated in the form of a compensation parameter vector. For example, the compensation parameter vector may include at least one of a fuel flow rate correction value, a GV angle correction value, a combustor fuel distribution ratio correction value, a temperature scale correction value, and other control parameter correction values.

[0122] Next, a parameter compensation application step (S190) may be performed. In the parameter compensation application step (S190), the correction parameter generation unit or the FBD dynamic compensation module may calculate a new control parameter by reflecting an optimal compensation amount to the existing control parameter. For example, the new control parameter may be calculated by adding an optimal compensation amount to the existing control parameter.

[0123] Additionally, in the parameter compensation application step (S190), the FBD dynamic compensation module can reflect the calculated compensation parameters in the FBD controller. For example, the FBD dynamic compensation module can update at least one of the fuel flow rate command, GV angle command, TIT scale, fuel distribution mapping, or combustor fuel distribution ratio mapping. Furthermore, the AI ​​compensation signal can be injected into the existing FBD control logic via a feedforward or feedback compensation method, and the compensated control signal can be applied to the gas turbine.

[0124] Next, a performance verification step (S200) may be performed. In the performance verification step (S200), the actual output value after the corrected control parameter is applied may be measured. Subsequently, the output deviation calculation unit may calculate a new output deviation and a normalized deviation by comparing the actual output value after correction with the theoretical output value again.

[0125] Next, a step for determining whether performance criteria are satisfied (S210) may be performed. In the step for determining whether performance criteria are satisfied (S210), if the new normalized deviation is below a preset success criterion, it may be determined that the correction parameter has been appropriately applied. Conversely, if the new normalized deviation exceeds the success criterion, the correction parameter generation unit may return to the optimal compensation amount calculation step (S170) to recalculate the compensation amount. Accordingly, the parameter compensation algorithm of the present invention can gradually reduce the output deviation through real-time iterative execution.

[0126] Next, a training data update step (S220) may be performed. In the training data update step (S220), at least one of driving data, output deviation, correction parameters, correction application results, and performance verification results may be stored as training data. The stored data may be used for retraining a PINN-based virtual sensor model or a PHM model.

[0127] Next, a model enhancement step (S230) may be performed. In the model enhancement step (S230), a PINN-based virtual sensor model or a PHM model may be retrained using stored training data. For example, model retraining may be performed periodically or in a conditional manner when the output deviation accumulates above a reference value. Additionally, in the model enhancement step (S230), highly reliable operating data may be selected and reflected in the model training. Accordingly, the prediction accuracy and generalization performance of the virtual sensor model may be improved, and adaptability to the aging of the gas turbine due to long-term operation or changes in operating conditions may be improved. Meanwhile, as illustrated in FIG. 4, the parameter compensation algorithm of the present invention may be performed in conjunction with FBD control logic. For example, the FBD control logic may include at least one of a load controller, a fuel flow controller, a combustor temperature controller, a fuel distribution mapping, a GV angle mapping, and a combustor fuel distribution ratio mapping. The present invention can improve control performance by injecting correction parameters calculated based on output deviation into the existing control logic without replacing the existing FBD control logic.

[0128] Meanwhile, the methods according to the various embodiments of the present invention described above can be implemented in the form of an application or software program that can be installed on an existing electronic device.

[0129] In addition, the whole or part of the method may be composed of multiple software function modules and implemented on an operating system (OS). Alternatively, each step may be composed of a single software function module, or each step may be combined to form a single software function module and implemented on an operating system. Therefore, even if all of the embodiments of the present disclosure are not implemented as a single software function module, if multiple software function modules implement each step of the present disclosure and multiple software function modules are implemented on a single operating system, it can be understood that the method of the present disclosure has been implemented.

[0130] In addition, the methods according to the various embodiments of the present invention described above can be implemented solely through software upgrades or hardware upgrades of existing electronic devices. Furthermore, the various embodiments of the present invention described above can also be performed through an embedded server equipped in an electronic device or an external server of the electronic device.

[0131] Meanwhile, according to one embodiment of the present invention, the various embodiments described above may be implemented as software comprising instructions stored on a computer-readable recording medium using software, hardware, or a combination thereof. In some cases, the embodiments described herein may be implemented as the processor itself. According to the software implementation, embodiments such as the procedures and functions described herein may be implemented as separate software modules. Each of the software modules may perform one or more functions and operations described herein.

[0132] Meanwhile, a computer or a similar device may include a device according to the disclosed embodiments, which is capable of calling instructions stored from a storage medium and operating according to the called instructions. When said instructions are executed by a processor, the processor may perform a function corresponding to said instructions directly or by using other components under the control of said processor. The instructions may include code generated or executed by a compiler or an interpreter.

[0133] A computer-readable recording medium may be provided in the form of a non-transitory computer-readable recording medium. Here, "non-transitory" simply means that the storage medium does not contain a signal and is tangible, without distinguishing whether data is stored semi-permanently or temporarily on the storage medium. In this context, a non-transitory computer-readable medium refers to a medium that stores data semi-permanently and is readable by a device, rather than a medium that stores data for a short moment, such as registers, caches, or memory. Specific examples of non-transitory computer-readable media may include CDs, DVDs, hard disks, Blu-ray discs, USBs, memory cards, and ROMs.

[0134] As described above, exemplary embodiments have been disclosed in the drawings and specification. Although specific terms have been used to describe the embodiments in this specification, they are used only for the purpose of explaining the technical concept of this disclosure and are not intended to limit the meaning or the scope of this disclosure as defined in the claims. Therefore, those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. Accordingly, the true technical scope of protection of this disclosure should be determined by the technical concept of the appended claims. Explanation of the symbols

[0135] 110: Preprocessing section 120: Virtual sensor unit 130: Deviation calculation unit 140: Correction parameter generation unit 150: Safety Constraints and Reliability Assessment Department 160: FBD Dynamic Compensation Module 310: Measurement and Input Signal Section 320: FBD-based gas turbine control logic 330: AI Correction Module 340: Gas turbine 350: AI Correction Signal Injection Unit 360: Actuator

Claims

Claim 1 A turbine control system based on a physics-compliant artificial neural network, comprising: a data collection and preprocessing unit (110) for collecting and preprocessing operating data of a gas turbine; a virtual sensor unit (120) for estimating thermodynamic state variables inside the turbine using a physics-compliant artificial neural network (PINN) learned based on a loss function including physics laws, with the preprocessed operating data as input; an output deviation calculation unit (130) for calculating a theoretical output value under current operating conditions based on the estimated thermodynamic state variables and calculating an output deviation between the theoretical output value and the actual output value; a correction parameter generation unit (140) for generating control correction parameters to correct control parameters of an existing turbine controller based on the output deviation; a safety constraint and reliability evaluation unit (150) for evaluating whether the control correction parameters are applicable; and an FBD dynamic compensation module (160) for controlling the gas turbine by reflecting the control correction parameters in a functional block diagram (FBD)-based control logic. Claim 2 A turbine control system according to claim 1, wherein the data collection and preprocessing unit (110) collects at least one of the gas pressure, gas temperature, gas flow rate, and generator output data of the gas turbine in real time from a turbine / generator sensor (instrument) (100), and performs at least one of normalization, noise removal, and outlier correction on the collected operation data. Claim 3 A turbine control system according to claim 1, wherein the above-mentioned physical law-compliant artificial neural network is included in the above-mentioned virtual sensor unit (120) and is trained based on a loss function comprising at least one of a data loss term corresponding to an error between the driving data and the predicted value, a physical loss term reflecting the energy conservation law of the gas turbine, a constraint term reflecting entropy change conditions, and a boundary loss term reflecting input / output boundary conditions. Claim 4 A turbine control system according to claim 1, wherein the virtual sensor unit (120) estimates at least one of the enthalpy of a compressor or combustor, compressor efficiency, combustor state, turbine efficiency, and change in entropy, which is difficult to measure directly, using the physical law-compliant artificial neural network. Claim 5 A turbine control system according to claim 1, wherein the output deviation calculation unit (130) analyzes at least one of the time response characteristics and steady-state error of the output deviation and classifies the cause of the output deviation as either an improper control parameter or a degradation of equipment performance based on the analysis result. Claim 6 A turbine control system according to claim 1, wherein the correction parameter generating unit (140) generates at least one of a feedforward correction value for a guide vane valve control signal, a proportional gain correction value for a PID controller, an integral gain correction value, and a derivative gain correction value as a control correction parameter based on the output deviation. Claim 7 A turbine control system according to claim 1, wherein the safety constraint and reliability evaluation unit (150) determines whether the control correction parameter is within a preset safety range and evaluates the output reliability of the physics law-compliant artificial neural network based on the residual of the physics equation of the physics law-compliant artificial neural network or the abnormality of the input data, and if the control correction parameter is outside the safety range or the output reliability is below a preset threshold, stops the application of the control correction parameter and returns to the existing FBD-based control logic. Claim 8 A method for controlling a gas turbine, comprising: a step of collecting and preprocessing operating data of a gas turbine (S110); a step of inputting the preprocessed operating data into a physics-compliant artificial neural network (PINN) to estimate thermodynamic state variables inside the turbine (S120); a step of calculating a theoretical output value under current operating conditions based on the estimated thermodynamic state variables and calculating an output deviation between the theoretical output value and the actual output value (S130); a step of generating a control correction parameter to correct the control parameters of an existing turbine controller based on the output deviation (S140); a step of evaluating whether the control correction parameter is applicable (S150); and a step of controlling the gas turbine by reflecting the control correction parameter in a functional block diagram (FBD)-based control logic (S160). Claim 9 A turbine control method according to claim 8, wherein the step of collecting and preprocessing the above-mentioned operating data (S110) comprises: a step of collecting at least one of the gas pressure, gas temperature, gas flow rate, and generator output data of the above-mentioned gas turbine in real time (S111); and a step of performing at least one of normalization, noise removal, and outlier correction on the collected operating data (S112). Claim 10 A turbine control method according to claim 8, wherein the above-mentioned physical law-compliant artificial neural network is trained based on a loss function comprising at least one of a data loss term corresponding to an error between the operating data and the predicted value, a physical loss term reflecting the energy conservation law of the gas turbine, a constraint term reflecting an entropy change condition, and a boundary loss term reflecting an input / output boundary condition. Claim 11 In claim 8, the step of estimating the thermodynamic state variable (S120) is characterized by estimating at least one of the compressor enthalpy, combustor enthalpy, compressor efficiency, combustor state, turbine efficiency, and entropy change amount, which are difficult to measure directly, using the physical law-compliant artificial neural network. Claim 12 A turbine control method according to claim 8, wherein the step (S130) of calculating the output deviation analyzes at least one of the time response characteristics and steady-state error of the output deviation, and classifies the cause of the output deviation as either inappropriateness of control parameters or degradation of equipment performance based on the analysis result. Claim 13 A turbine control method according to claim 8, wherein the step (S140) of generating the control correction parameter is characterized by generating at least one of a feedforward correction value for a guide vane valve control signal, a proportional gain correction value for a PID controller, an integral gain correction value, and a derivative gain correction value as the control correction parameter based on the output deviation. Claim 14 A turbine control method according to claim 8, wherein the step (S150) of evaluating the applicability of the control correction parameter determines whether the control correction parameter is within a preset safety range, evaluates the output reliability of the physics-law-compliant artificial neural network based on the residual of the physics equation of the physics-law-compliant artificial neural network or the abnormality of the input data, and, if the control correction parameter is outside the safety range or the output reliability is below a preset threshold, stops the application of the control correction parameter and returns to the existing FBD-based control logic.

Images