Control and tuning of gas turbine combustion

Abstract

The present invention is entitled "Control and Tuning of Gas Turbine Combustion". The present invention discloses a system, the system comprising: a gas turbine having a combustion system; a control system; and a combustion auto-tuner communicatively coupled to the control system, the combustion auto-tuner comprising: an optimization system having an empirical model and an optimizer of the combustion system; and sensors configured to measure the input and the output of the combustion system. The tuning process comprises the following steps: receiving current measurement results from the sensors; using the optimization system to calculate an optimized control scheme based on the current measurement results received from the sensors; and transmitting the optimized control scheme to the control system.

Description

Technical Field

[0001] The present invention relates to a power system and, in particular, to an automated system for sensing operating conditions within a combustion system of a gas turbine and adjusting or tuning the combustion system to achieve desired operation or performance. Background Art

[0002] Industrial combustion or gas turbines are often used in many applications, such as for driving generators in power plants. Many of these gas turbines achieve low emission levels, such as NOx emissions, by using lean premixed combustion. In lean premixed combustion, fuel and excess air are mixed before combustion to control and limit thermal NOx production. Such combustion systems or burners (commonly referred to as dry low NOx burners) require careful management of combustion conditions to achieve stable operation and acceptable NOx and CO emissions while maintaining no pressure oscillations (referred to as "burner dynamics" or simply "dynamics"), which typically include a combination of sound and unstable energy release during the combustion process. Such systems typically require multiple independently controlled fuel injection points provided by multiple fuel circuits or fuel nozzles in each of one or more parallel burners to achieve desired combustion performance within a range of operating modes (e.g., between startup and full load). Although these combustion systems show great benefits from an emissions perspective, the operating range of these systems is greatly reduced compared to more conventional combustion systems. As a result, control of fuel conditions and delivery, such as how the fuel is distributed and injected into the combustion zone through the various nozzles, also known as "fuel splitting," has become a critical operating parameter and often requires frequent adjustments, for example, when ambient atmospheric conditions (such as temperature, humidity, and pressure) change, when the gas turbine degrades with use, when the fuel composition changes, or when other variables are introduced. Adjusting combustion fuel conditions, distribution, and injection to account for these changes is called "combustion tuning" or simply "tuning."

[0003] Controlled operation of the combustion system typically employs manually setting operating parameters associated with the combustion system of the gas turbine under several expected or standard operating conditions. Such adjustments or tuning of the combustion system typically require technicians to spend a day or more using specialized instruments to complete. However, although the adjusted settings may be satisfactory when set, changing conditions (such as changes in environmental conditions, changes in equipment conditions, and unexpected operating conditions) may cause unacceptable performance within a few days or weeks, which may require costly manual tuning again. Some of these conventional methods may use algorithms and / or physics-based models to predict emissions and other outputs based on current gas turbine operating parameters, and thereby select or modify set points, such as set points associated with fuel distribution or total fuel / air ratio. However, these methods do not utilize real-time data related to combustor dynamics and emissions, which results in dynamic increases and unexpected emission excursions within the combustion system caused by the control scheme generated by the prediction. Pressure pulsations associated with such dynamics may have sufficient force to damage or degrade the combustion system and significantly reduce the life of the combustion hardware, and such emission excursions may lead to costly permit violations.

[0004] Therefore, systems and methods for promoting combustion stability, reducing dynamics and / or improving emission levels remain a needed and highly interesting area of ​​technology in the industry. In addition, systems and methods for optimizing combustion system performance by operating with near real-time data obtained from turbine sensors to adjust fuel splits, fuel temperature, overall fuel / air ratio and / or other variables would be of significant value. Summary of the invention

[0005] Thus, the present application describes a system comprising: a gas turbine having a combustion system; a control system operably connected to the gas turbine to control the operation of the turbine; and a combustion auto-tuner communicatively coupled to the control system, the combustion auto-tuner comprising: an optimization system having an empirical model of the combustion system and an optimizer; sensors configured to measure inputs and outputs of the combustion system; a hardware processor; and a machine-readable storage medium having instructions stored thereon that cause the hardware processor to execute a tuning program to tune the operation of the combustion system. The tuning program comprises the following steps: receiving current measurements of the inputs and outputs from the sensors; calculating an optimized control scheme for the combustion system using the optimization system in view of the current measurements received from the sensors; and transmitting the optimized control scheme to the control system.

[0006] These and other features of the present application will become apparent upon review of the following detailed description of the preferred embodiments, taken in conjunction with the accompanying drawings and the appended claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] These and other features of the present invention will be more fully understood and appreciated through careful study of the following more detailed description of exemplary embodiments of the present invention taken in conjunction with the accompanying drawings, in which:

[0008] Figure 1 Schematic diagrams of power systems according to aspects of the present disclosure or within which embodiments of the present disclosure may be used are shown.

[0009] Figure 2 Schematic diagrams of exemplary distributed computing networks and power plants with gas turbines are provided, including computer-implemented control systems and network resources according to embodiments of the present disclosure or illustrating an environment within which such embodiments may be enabled or implemented.

[0010] Figure 3 A simplified representation of a system including a gas turbine according to an embodiment of the present disclosure is shown.

[0011] Figure 4 A simplified representation of a combustion system according to an embodiment of the present disclosure is shown.

[0012] Figure 5 A method for determining a fuel split for a gas turbine is shown using a conventional fuel split system.

[0013] Figure 6 An exemplary combustion auto-tuner according to an embodiment of the present disclosure is shown.

[0014] Figure 7 A system for monitoring combustor dynamics according to an embodiment of the present disclosure is shown.

[0015] Figure 8 An alternative combustion auto-tuner according to an embodiment of the present disclosure is shown.

[0016] Fig. 9 A block diagram is provided illustrating aspects of an optimization system according to an exemplary embodiment of the present disclosure.

[0017] Fig.10 A model with disturbance rejection according to the present disclosure is shown that can be used with an optimization system for a combustion auto-tuner.

[0018] Fig.11 A model with additional fixed bias settings is shown that may be used with an optimization system for a combustion autotuner in accordance with the present disclosure.

[0019] Fig.12 Shows Fig.10 A simplified version of the model.

[0020] Fig.13An aggregate model that may be used with an optimization system for a combustion auto-tuner according to the present disclosure is shown.

[0021] Fig.14 An example “experimental design” for collecting data to train a model in accordance with the present disclosure is shown.

[0022] Fig.15 A method for automatic learning of a model according to the present disclosure is shown.

[0023] Fig.16 An exemplary learning run for acquiring a training data set to train a model according to the present disclosure is shown. DETAILED DESCRIPTION

[0024] Exemplary embodiments will be described more fully below with reference to the accompanying drawings, in which some but not all embodiments are shown. In fact, the embodiments of the present disclosure may take many different forms and should not be construed as being limited to the embodiments listed herein. Like numbers may always refer to like elements.

[0025] Figure 1 A schematic diagram of an electric power system 10 is shown, which illustrates an exemplary electric power system environment in which embodiments of the present disclosure may operate. The electric power system 10 includes several power plants 12 for generating electric power. Such power plants 12 may include wind power plants and thermal power plants 12 as shown, and may also include other types of power plants, such as solar power plants, hydroelectric power plants, geothermal power plants, or nuclear power plants. Within the electric power system 10, a public transmission line 14 connects the power plant 12 to one or more loads or users 16, which may include, for example, municipalities, residential and industrial users. The transmission line 14 represents a distribution network or power grid of the electric power system 10, and may include multiple sections or substations as appropriate.

[0026] The power system 10 also includes a control system or controller that manages or controls the operation of several components included therein. For example, a power plant controller 22 can control the operation of the power plant 12 and the generator sets included therein, while a load controller 23 can control the manner or time of power purchase by the user 16. The power system 10 also includes a central management agency or dispatch management agency 24 that manages the dispatch process, through which load commitments are allocated between the power plants 10 to meet user needs. The controllers 22, 23 can be connected to a communication network 20 on which data is exchanged via communication lines. The communication network 20 can be connected to or part of a communication network, such as the Internet, a private network, or a cloud network. In addition, the controllers 22, 23 can receive information, data, and instructions from databases and resources (or "data resources 21") through the communication network 20, and / or send information, data, and instructions to databases and resources (or "data resources 21"). The controllers 22, 23 can also store or accommodate data repositories locally. The data resources 21 may include several types of data, including but not limited to market data, operational data, performance data, and environmental condition data.

[0027] For example, in operation, the power plants 12 generate electricity, which is distributed over the utility transmission lines 14 for delivery to the users 16. According to the dispatch process managed by the dispatch management agency 24, the power plants 12 compete with each other to determine how each power plant will participate in future power generation periods to meet the expected load requirements of the users 16. In the dispatch process, the operators of the power plants 12 generate bid curves, which include the cost of generating electricity for future power generation periods. The bid curve represents the bids of the power plants 12 for future power generation periods, and generally includes an incremental variable cost curve or some other suitable indication of the variable generation cost, which can be expressed, for example, in dollars per megawatt-hour relative to megawatt output. The dispatch management agency 24 then uses the received bid curves to allocate the expected load requirements for the future power generation period among the power plants 12. For example, the dispatch management agency 24 may use a competitive process called economic dispatch to determine the level of participation of each power plant 12 that most effectively meets the predicted load requirements of the power system 10. In this way, the main goal of the dispatch management agency 24 is to find the lowest cost of generating electricity to meet the expected user needs. Once completed, the dispatch authority 24 issues a commitment schedule to the generating stations 12 that specifically describes how each generating station will participate in future generation periods.

[0028] Once the commitment schedule is communicated to the power plant 12, the goal of each power plant is to produce the promised output in a way that maximizes economic returns. It should be understood that this goal is becoming particularly challenging given the increasing complexity of modern power plants. This complexity is the result of a combination of factors. For example, power plants now typically include many different types of different generator sets, and these can achieve many optional power generation configurations, each of which has its own set of economic considerations. Each of these generator sets must also be maintained according to its own maintenance schedule, which requires regular interruptions to the power supply, which must be carefully planned to avoid unnecessary impacts on power plant operations. In addition, the power industry is a very formal industry, and there are many laws and regulations that affect the operation of power plants. Finally, of course, volatile market conditions make short-term and long-term profitability a constantly changing goal. Although conventional control systems can skillfully solve some problems, such as adjusting fuel flow, inlet guide vane settings and other inputs by executing scheduling algorithms, the actual output and efficiency of power plants are often affected by external factors that cannot be fully anticipated (such as variable environmental conditions or unplanned outages). It should be appreciated that the increasing complexity of power plants and the variability of operating conditions result in difficulty in predicting and controlling performance, which often results in inefficient operation. Furthermore, conventional control systems are often inadequate because they do not fully exploit the growing data-intensive aspects of the industry. Specifically, current control systems have not successfully achieved the level of industrial optimization brought about by the increasing availability of operating data, and this failure has resulted in power plants and generator sets operating inefficiently and without maximizing economic returns.

[0029] Additionally, performance degradation is another factor that is difficult to quantify, but can have a significant impact on power plant operations and economic returns. More specifically, because the performance of a power plant's generator sets and machinery degrades over time, this must be taken into account when scheduling the units and understood when evaluating the cost-effectiveness of the power plant. However, the rate at which a gas turbine degrades varies depending on many factors, including operating hours, load, operating configuration, transient operation, and number of starts. However, many of these factors are interrelated and affect degradation in ways that are not fully understood. For example, the life hours of a gas turbine can be extended by reducing the combustion temperature, however, this alternative reduces efficiency and increases fuel costs. Conversely, increasing the combustion temperature increases efficiency but shortens component life and increases maintenance and part replacement costs. Similarly, decisions related to turndowns, shutdowns, and rapid ramping affect the life consumption of machine components and fuel costs. Furthermore, by controlling a power plant's generator sets in an extremely static manner, i.e., by using static control profiles such as heat rate curves derived from periodic performance testing, grid compliance and dispatch can be adversely affected. Between these updates, unit performance may have degraded to the point where the control profiles are out of date.

[0030] To compensate for all this variability, power plant operators tend to be extremely conservative when planning for future generation periods, which leads to bidding practices that make their units less competitive and underutilize capacity. Otherwise, there is a risk that the power plant will bid beyond its current performance capabilities, which may force the operator to operate the units inefficiently to meet load commitments. As a result, operators continue to struggle to manage the life cycle costs and economic returns associated with a power plant's generating units due to many complex and interrelated factors. The degree to which these factors can be better understood and the associated trade-offs can be more effectively balanced can directly translate into increased economic benefits for the power plant. That is, without understanding and identifying short-term inefficiencies and long-term degradation, conventional power plant control systems must rely on frequent re-tuning, which can be costly, or bid and operate conservatively, which often has a negative impact on the cost-effectiveness of the power plant.

[0031] Integrated software- and hardware-based controllers have long been the preferred solution for driving the operation of industrial assets and machines, such as those within power plants 12. However, with the rise of cheap cloud computing, the increase in sensor functionality and the decrease in sensor costs, and the proliferation of mobile technology and networking capabilities, new possibilities have emerged to transform the way power plants and generator sets are designed, operated, and maintained. For example, recent advances in sensor technology now enable the collection of new types and more operational data, while advances in network speeds and capacity allow that data to be transmitted to remote locations in essentially real time. This means that, for example, even for fleets of similar assets that are geographically dispersed, the data collected at each remote site can be effectively aggregated, analyzed, and used in a manner designed to improve the performance of the fleet and individual assets. As a result of this evolving and data-intensive environment, new opportunities have emerged to improve or optimize the value of industrial assets through new, industry-centric hardware and software solutions. However, conventional control systems have failed to achieve the levels of industrial asset optimization that are made possible by the increased amount of new types of available operational data.

[0032] Together, these problems often result in industrial assets, such as power plants and generator sets, being operated in a manner that does not maximize efficiency and economic return. Embodiments of the present disclosure will now be disclosed with reference to the following figures, which address one or more aspects related to these persistent problems.

[0033] Figure 2 A schematic diagram of an exemplary distributed computing system or network within an industrial environment is provided, wherein one exemplary power plant type, namely a power plant 25, has one or more generator sets 26. As will be seen, in accordance with embodiments of the present disclosure or illustrating a networked industrial environment in which such embodiments may be enabled or implemented, the power plant 25 and the generator sets 26 may be networked to control systems, controllers, and other resources via internal and / or external networks. Figure 2 Also presented is an exemplary system by which operational data relating to any industrial asset can be collected and then analyzed and used for control and management of the industrial asset. As will be seen, such operational data can be used to create and train empirical models through a machine learning process or a deep learning process.

[0034] As an exemplary type of generator set 26, the power plant 25 is shown as including a combustion or gas turbine 27, however it should be understood that it may include other generator sets 26 of different types. The generator set 26 can be monitored by a variety of sensors that collect data and transmit the data to local and remote computer systems for analysis by the system and control and operational feedback generated by the system. The gas turbine 27 may include a compressor 28 coupled to a downstream turbine section or turbine 29 via a common shaft, with a combustor 30 positioned between the two. In an operating example, the rotation of the rotor blades within the compressor 28 compresses the air flow. Within the combustor 30, when the compressed air is mixed with fuel and burned, energy is released, and the resulting hot gas flow or working fluid is directed to the rotor blades within the turbine 29, causing the shaft to rotate. In this way, the energy in the working fluid is converted to rotate the shaft, which can then be used, for example, to drive the coils of the generator 35 to generate electricity.

[0035] The power plant 25 may include any number of subsystems, components, generator sets 26, or industrial assets, each of which may be connected to a computer-implemented control system and communication network for control and performance optimization in accordance with the analysis, processes, and other functions described herein. For example, within such a control system, operational data may be collected and provided electronically to computer-implemented analysis, data repositories, power plant or unit controllers, cloud-based services, and / or user interfaces or input devices (or "user devices") to enable a variety of functions. These functions may include deriving control set points for operating industrial assets, data storage / calling, generating graphical user interfaces, prompting / obtaining user input, decision support for long-term and short-term operations, automatic control responses, etc., as may be provided herein. Therefore, Figure 2 The networked computing systems, controllers, and associated computing resources can send and receive electronic communications to and from each of the systems, subsystems, components, sensors, actuators, user devices, and / or other devices within the power station 25 and the included generator sets 26, and can regulate control settings associated therewith in a manner consistent with the functionality described herein, whether such functionality is explicitly stated or implied in the appended claims, as understood by a person of ordinary skill in the art.

[0036] As shown, each of the generator sets 26 may include a control system or controller (or "set controller 39") that monitors and controls operations. The set controller 39 is associated with controlling the gas turbine, and may also be referred to as a gas turbine controller or "GT controller", and may implement a variety of control requirements and prevent adverse or abnormal operating conditions. For example, the set controller 39 may perform many functions, including fuel, air and emission control, fuel sequencing for startup, shutdown and cooldown, synchronization and voltage matching of the generator 35, monitoring gas turbine control and auxiliary functions, preventing unsafe and adverse operating conditions, and other functions, as may be described or implied herein. A power plant controller 40 may also be provided that communicates with each of the set controllers 39. The power plant controller 40 generally controls aspects related to the overall operation of the power plant 25. The power plant controller 40 may also be connected to other power plant controllers 40 associated with the operation of other power plants 25, so that multiple power plants 25 within a power plant fleet can communicate, exchange information, and control as a group.

[0037] The unit controller 39 and the power plant controller 40 (also collectively referred to herein as "controllers") may each include a computer system having a digital processor or processing resources (or "processor") and a machine-readable storage medium or storage capacity (or "memory"). Alternatively, the unit controller 39 and the power plant controller 40 may be combined into a single controller with an integrated architecture. The unit controller 39, the power plant controller 40, and the computer system associated with each controller may be connected to one or more user devices 44. As shown, such connection may be made through an internal and / or external network. The user device 44 may be configured to receive / send communications from / to any person associated with the operation of the power plant 25 or the generator set 26. The user device 44 may include any conventional device with a user interface, including but not limited to mobile devices, laptops, and other workstations, whether the location relative to the power plant 25 or the generator set 26 is local or remote.

[0038] The computer systems associated with controllers 39, 40 may each include one or more processors, memory, and other conventional computing components as required for any of the functions described herein. For example, the computer systems of controllers 39, 40 may include distributed Figure 2 Non-local aspects of several other resources, devices, or nodes shown. Such nodes may include, but are not limited to, a communication network, cloud, or cloud-based platform (or, as shown, "cloud network 48" or more generally, "the cloud"); a data repository or other information resource (or, as shown, a "data resource" or "data resource 21"); several user devices 44; and other fleet controllers 39, power plant controllers 40, and industrial assets 49, as necessary to implement any of the functionality described herein. It should be understood that unless otherwise stated, Figure 1The unified and partial depictions of controllers 39, 40 and computer systems, processors, and memory resources associated with each controller are exemplary only and are not intended to limit the manner in which any of the described functions are performed. Along these lines, it should be further understood that references to unit controllers 39, plant controllers 40, user devices 44, and / or computer systems associated with each may include computing resources such as processing, storage, memory, data, and communication equipment distributed and accessed via an internal or plant-level network and through a cloud network 48, as should be understood by those of ordinary skill in the art given the nature and functionality of the state-of-the-art distributed computing environment described herein, whether explicitly stated or implied.

[0039] The gas turbine 27, as well as any of the other generator sets 26, may include a plurality of sensors 56 configured to monitor specific operational aspects of the gas turbine 27 by detecting or measuring operating conditions or parameters of the entire engine while the engine is running. For example, the sensors 56 may include temperature sensors, pressure sensors, speed sensors, flame detector sensors, valve position sensors, and any other conventional sensors expected given the functionality described herein. As used herein, the term "parameter" refers to a measurable physical characteristic of an operation that can be used collectively to define operating conditions within a system. These operating parameters may include, but are not limited to, temperature, pressure, humidity, gas flow characteristics, environmental conditions, fuel characteristics, and other measurable characteristics that may be described or implied by the functionality described herein.

[0040] The gas turbine 27, as well as any other generator set 26, may include several actuators 57 through which control of the gas turbine 27 is achieved. For example, the actuator 57 may include an electromechanical device with a variable set point or setting that allows manipulation of certain process inputs (referred to herein as "manipulated variables") for controlling process outputs (referred to herein as "controlled variables") according to a desired result or mode of operation. Control of the manipulated variables by the actuators 57 must take into account other uncontrollable variables, referred to herein as "disturbance variables", which also affect the operation of the gas turbine 27. Thus, for example, a command generated by the controller 39, 40 may cause one or more actuators 57 within the gas turbine 27 to adjust a valve between the fuel supply and the combustor 30 to adjust a flow level, fuel split, or fuel type.

[0041] It should be understood that Figure 2A computer system, such as controller 39, 40, executes code or a software program or application (or, generally referred to as "software") that is configured to control the generator set 26 and / or power plant 25 according to a desired operating mode, or to maximize a predefined performance goal. Such control may be responsive to operating data provided by sensors 56 and / or instructions received from user equipment 44, and may be achieved by manipulating one or more actuators 57. To facilitate this, user equipment 44 may be accessed and used by power plant managers, technicians, engineers, operators, energy traders, owners, and / or other stakeholders, as may be described or implied by any functionality provided herein. The software may include schedules, analyses, models, and algorithms for regulating any system or subsystem described herein.

[0042] Generally, the cloud network 48 includes an external network or wide area network that connects the power plant 25 to other power plants 25, industrial assets 49, and data resources 21. A more secure local area network or intranet for connecting local resources may also be provided. Figure 2 The connection established between the cloud network 48 and any of the other nodes or devices may be wired or wireless and / or include any prior art communication system or device. It should be understood that the cloud network 48 may be part of a larger communication system or network, including the Internet and / or one or more private computer networks. In this way, the controllers 39, 40 may receive information, data, and instructions from the data resource 21 and / or may send information, data, and instructions to the data resource. A connection to these data resources 21 may be established via the cloud network 48, or, alternatively, one or more of the data resources 21 may be stored locally and one or more of the data resources 21 may be accessed outside the cloud network 48 via a private or internal network. In addition, as discussed in more detail below, the cloud network 48 may connect the gas turbine 27 to other industrial assets 49, such as other remote gas turbines, power plants, or facilities of suppliers or customers.

[0043] Data resources 21 may include any type of data implied by the functionality described herein. For example, data resources 21 may include, but are not limited to, the following types of data: market data, operational data, and environmental condition data. Market data may, for example, include information about market conditions, such as energy sales prices, fuel costs, labor costs, regulations, etc. Operational data may, for example, include information related to the operating conditions of the power station 25, gas turbine 27, or related components. Such operational data may include temperature or pressure measurements within the gas turbine 27, air flow rates, fuel flow rates, etc. Environmental condition data may, for example, include information related to environmental conditions at the power station 25, such as ambient air temperature, humidity, and / or pressure. Market, operational, and environmental condition data may each include historical records, current condition data, and / or data related to predictions. For example, data resources 21 may include: current weather / climate information and forecasted weather / climate information; current market conditions and forecasted market conditions; usage and performance history regarding the operation of power plant 25 or gas turbine 27; and / or, measured parameters regarding the operation of other similarly located power plants or gas turbines, which may be defined as those with similar components and / or configurations. Other data that may describe or imply the functionality described herein may also be stored in data resources 21 and called from there as needed.

[0044] Thus, according to an exemplary embodiment, it should be understood that although each of the controllers 39, 40 may include a computer system having a processor, memory, database, communication equipment and other computing resources, these resources may be distributed among, for example, Figure 2 , including controllers 39, 40, user devices 44, cloud networks 48, data resources 21, other industrial assets 49, etc. Although some aspects of the controllers 39, 40 may be set locally, other aspects may be remote and these other aspects are operatively connected via the cloud network 48, such as Figure 1As indicated by the dashed line. As described above, the controllers 39, 40 may be directly or indirectly connected to each device, subsystem or component within the power plant, including the gas turbine 27 and related subsystems, so that the sensor 56 delivers data to the controller and the actuator 57 responds to commands received from the controller. However, unless otherwise stated, the systems and methods of the present invention may include embodiments that are not configured to directly control the gas turbine or other generator set and / or directly measure operating conditions. In those embodiments currently described that directly measure / control the operation of the gas turbine 27 or other generator set 26, such control may be implemented according to software or a "control program", which may be stored in whole or in part in the memory of the computer system of the unit controller 39 and / or the power plant controller 40, and when running, the software or "control program" causes its computer system to manage data and execute the processes, analysis and logic provided herein.

[0045] According to an exemplary mode of operation, the processor of the computer system of the controller executes software or program code (referred to herein as "program code") defining the control program. When executing the program code, the processor processes data, which enables the converted data to be read from the memory and / or written to the memory. The display on the user device 44 enables a human user (also referred to as an "end user" or "operator"), such as any of those users described herein, to interact with any of the computer systems using a communication link such as a cloud network 48. It should be understood that the cloud network 48 enables the computer system to communicate with any other device described herein, regardless of their location. In this regard, the control program of the present disclosure can manage a set of interfaces that enable several users to interact with the control program. In addition, as further discussed below, the control program can manage (e.g., store, retrieve, create, manipulate, organize, present, etc.) data, such as control data or operational data. The controller may include one or more general-purpose computing products that, once installed on the control program, can execute the program code of the control program. As used herein, it should be understood that "program code" means any collection of instructions in any language, code, or notation that causes a computing device with information processing capabilities to perform a specific action directly or after any combination of the following: (a) conversion to another language, code, or notation; (b) copying in a different material form; and / or (c) decompression. In addition, program code may include object code, source code, and / or executable code, and when on a computer-readable medium, may form part of a computer program product. It should be understood that the term "computer-readable medium" may include one or more of any type of tangible expression medium now known or later developed from which a computing device may perceive, copy, or otherwise transmit a copy of the program code.

[0046] It should be understood that when a computer executes program code, the computer becomes a device for implementing the present invention, and on a general-purpose microprocessor, a specific logic circuit is created by configuring the microprocessor with a computer code segment. For example, the technical effect of the executable instructions can be to implement a control method and / or system and / or computer program product that uses a model to enhance, expand or optimize the operating characteristics of an industrial asset to improve economic returns under a given set of constraints (such as environmental conditions, market conditions, performance parameters, life cycle costs, etc.). In addition to using current information, historical information and / or predicted information can also be used, and feedback loops can be established to more effectively and dynamically operate generator sets and / or power plants during fluctuating conditions. The computer code of the control program can be written into computer instructions that can be executed by the computer system of the controller. In this regard, the control program executed by the controller and / or other distributed computer resources can be embodied as any combination of system software and / or application software. In addition, a set of modules can be used to implement the control program. In this case, a module can enable a controller to perform a set of tasks used by the control program, and the module can be developed and / or implemented separately from other parts of the control program. It should be understood that when the computer system executing the control program includes multiple computing devices, such as previously described, each computing device may have only a portion of the control program or program code fixed thereon.

[0047] Thus, in general, the control program enables computing and digital resources (such as those specifically described herein, or those that may be generally referred to as "computer systems") to implement a unit controller or power plant controller according to the functions provided herein, particularly according to those diagrams including data flow diagrams, algorithms, methods, analysis and / or logic to be followed. For the purposes of this document, such a computer system may obtain data by any conventional means. For example, such a computer system may calculate control data for a generator set or power plant, retrieve control data related to a generator set or power plant from one or more data repositories, data storage libraries or data sources, and / or receive control data for a generator set or power plant from other systems or devices inside or outside the location of the generator set or power plant. In other embodiments, the present application may include a method for providing a copy of program code, such as for executing a control program, which may include implementing some or all of the processes described herein. It should be further understood that various aspects of the present disclosure may be implemented as part of a business method for performing the processes described herein on a subscription or payment basis. For example, a service provider may implement a control program at a customer's generator set or power plant. In this case, the service provider may manage a computer system or controller that performs one or more of the processes described herein for a customer facility.

[0048] Reference now Figure 3, a simplified representation of a gas turbine 27 is provided which will be used to describe certain aspects of the present disclosure. It should be understood that within the gas turbine 27, air supplied by a compressor 28 is combined with fuel within each combustor 30, wherein the air and fuel are inputs that control the amount of load generated by the turbine 29. As already described, characteristics of the air inflow, such as temperature, pressure, and humidity, may be measured at the inlet of the compressor 28 by one or more sensors. In addition, the air-fuel ratio of the resulting combustion may be controlled by controlling the inlet guide vanes ("IGV") of the air flow within the compressor 28. The exhaust gas of the gas turbine 27 contains emissions such as nitrogen oxides ("NOx") and carbon monoxide ("CO"), and after leaving the turbine 29, the exhaust gas is directed through a chimney 105. Within the chimney 105, the levels of such emissions may be measured by sensors 107 that are part of a continuous emissions monitoring system ("CEMS") 108. The CEMS 108 may transmit those measurements to a control system or controller (or "GT controller") of the gas turbine. The GT controller, for example, may be the one described above with respect to Figure 2 The unit controller 39 in question.

[0049] refer to Figure 4 , the gas system 109 is shown relative to the combustor 30 of the gas turbine. It should be understood that the manner in which the fuel and air are concentrated within the combustor 30 affects NOx emissions, CO emissions, and combustor dynamics, as well as lean blowout ("LBO"), flashback, and other characteristics of combustion performance. Therefore, the air and fuel supply can be delivered by splitting, so that the delivery to the combustor 30 can be controllably changed. These splits may include an air splitter 111 and a fuel splitter 113. The air splitter 111 can deliver a variable portion of the combustor air supply to two or more different injection points within the combustor 30, for example, directing the variable portion as bypass air injected in a downstream area of ​​the combustion zone. That is, the air splitter 111 generally operates by changing the distribution of the air supply between one or more injection points (e.g., between one or more main injection points located near the front end or head end of the combustor 30 and a downstream injection point of the bypass air) in response to an air split command from the GT controller.

[0050] The fuel supply delivered to the combustor 30 may be divided by the fuel splitter 113 so that a varying amount may be sent to two or more different nozzles or nozzle types within the combustor 30. These different nozzles or nozzle types are generally referred to herein as the first nozzle, the second nozzle, and the third nozzle (if present). Unless otherwise specified, it should be understood that each of the first nozzle, the second nozzle, and the third nozzle may represent any type of nozzle or nozzle group, and each nozzle may be distinguished from the other nozzles according to the nozzle type, configuration, quantity, position, or other criteria. According to a typical combustor arrangement, for example, the first nozzle may represent one or more pilot nozzles, the second nozzle may represent one or more main nozzles, and the third nozzle may represent one or more delayed lean nozzles or downstream nozzles (which are used to inject fuel and / or air in the downstream area of ​​the combustion zone within the combustor 30 and / or the upstream portion of the turbine 29). The first nozzle, the second nozzle, and the third nozzle may include only fuel nozzles, fuel-air nozzles, or premixed fuel-air nozzles. The embodiments of the present disclosure may be used in combustors having only the first nozzle and the second nozzle, while other embodiments may be used in combustors having more than three nozzles or nozzle types. The fuel splitter 113 generally operates by varying the manner in which the combustor fuel supply is distributed among the first, second, and third nozzles of the combustor 30 in response to a fuel split command from the GT controller.

[0051] Still Figure 4 As shown, the combustion system 109 may include a combustor dynamic monitoring system ("CDMS") 115. For example, as part of the system of the present disclosure, the CDMS 115 may monitor various vibration responses or frequencies generated by the combustor 30 during operation via sensors 117. One or more alarm limits may be set, and if the measured frequency exceeds those limits, the CDMS 115 may issue an alarm. Alternatively, the measured frequency data and alarm limits may be sent to a GT controller, which may use these values ​​to determine whether the combustor 30 is operating normally or malfunctioning, and if a malfunction occurs, the GT controller may determine whether potential damage to the gas turbine may occur. If damage may occur, the GT controller may send an alarm communication or a trip of the gas turbine to avoid such damage. The sensors 117 used by the CDMS 115 may include pressure and / or acceleration sensors positioned on each combustor 30 for measuring frequency data. For example, the CDMS 115 may use a directly mounted or waveguide connected pressure or optical sensing probe to measure combustor dynamics.

[0052] Thus, in general, operational data related to the performance of the combustion system of the gas turbine may be collected by sensors and transmitted to the GT controller. Emission levels of exhaust gases may be measured by the CEMS within the stack, and those measurements are forwarded to the GT controller. Likewise, combustor dynamics may be measured by the CDMS 115 via pressure sensing probes located within the combustor, and those measurements are forwarded to the GT controller.

[0053] In addition, data related to the temperature and heating value of the fuel supply to the combustor may be collected via one or more conventional sensors and then transmitted to the GT controller. Since the combustion tuning process of the present disclosure may include adjusting the temperature of the fuel supply, the present embodiment may include two-way communication between the GT controller and the fuel heating unit, which is configured to heat the fuel before delivering it to the combustor. Specifically, data related to the temperature and heating value of the fuel may be measured and provided to the GT controller, which may issue a command to the fuel heating unit to increase / lower the fuel temperature as part of the tuning process. Similarly, other operating parameters described or proposed herein may also be measured by sensors and then transmitted to the control system of the gas turbine. Such operating data related to the gas turbine and the combustion system may be collected periodically, for example, once every second or every few seconds, or several times per minute, and the resulting data may be provided to the control system of the gas turbine. As will be seen, embodiments of the present invention may use this data stream to achieve real-time or periodic tuning and optimization of the combustion system of the gas turbine.

[0054] Reference now Figure 5 , a conventional fuel splitter 125 and known methods for calculating the fuel split will first be discussed. Conventionally, as shown, the fuel splitter 125 can calculate the fuel split as a function of the load percentage (as shown, "Load %"), that is, the load percentage is adjusted according to a "temperature compensation" value based on the compressor inlet temperature. Therefore, as shown in the upper curve 126, the current value of "Load %" can be used to initially determine the percentage of the fuel supply to be delivered to the pilot nozzle (or "Pilot %"), for example, at 90% load, the upper curve 126 shows that the fuel split for "Pilot %" is 12%. Then, the lower curve 127 is queried to determine whether the value will be adjusted due to the compressor inlet temperature. That is, using the lower curve 127, the calculated "Pilot %" can be adjusted based on the current value of the inlet temperature. For example, for an inlet temperature of 55°, the lower curve 127 shows that the value of "Temperature Compensation" is approximately negative 2 (-2). This value is applied to the "Pilot %" of 12% to give a final fuel split of 10% for the pilot nozzle. A value of 10% is then issued as a fuel split command.

[0055] Typically created or adjusted manually by a tuner or, as used herein, a "tuner". Figure 5 The fuel flow divider 125 uses curves 126, 127 Figure 5 The curves 126, 127 used by the fuel splitter 125 of the gas turbine are used. To do this, the tuner manually moves the gas turbine over the entire load range, and at predetermined load intervals, the tuner adjusts the fuel split to the pilot nozzle and observes its effect on emissions, combustor dynamics, etc. Based on these observations, the tuner then adjusts the two curves 126, 127 to reflect the preferred settings for that particular gas turbine. Typically, the "re-tuning" of the curves 126, 127 must be performed periodically, for example, every few weeks or months, each time requiring about 8 hours of manual work. It should be understood that the gas turbine must be re-tuned in this manner because internal changes and degradation occur as the gas turbine is fired or due to seasonal changes in environmental conditions. In order to reduce the need for manual re-tuning, certain known automatic tuning processes have been developed. However, conventional combustion automatic tuners have only achieved limited success. In addition, these conventional combustion automatic tuners are very expensive because they generally rely on first principles or physics-based models, which are very expensive to develop and maintain for each gas turbine frame. Therefore, there remains a need for a cost-effective empirical method for tuning gas turbines, and in particular, a method that is universally applicable to a variety of gas turbines of different frames and manufacturers.

[0056] Drawing on the systems and concepts introduced above, we now turn our attention to Figures 6 to 16, to further disclose the present invention through the description of one or more exemplary embodiments. As will be seen, according to the present disclosure, a system is provided for automatically tuning or optimizing fuel splits in a gas turbine to reduce emissions (e.g., NOx and CO) and prevent burner dynamic excursions. According to exemplary embodiments, the system of the present disclosure may formulate fuel split control as an optimization problem and use the results to control gas turbine operation. For example, the system may employ a data-driven or empirical model such as a neural network model in conjunction with an optimizer to solve for an optimized fuel split setting for a combustion system. The system may use such a model to predict the effects of changes in fuel splits on emissions and burner dynamics and optimize accordingly. As another aspect, the system may automatically manipulate fuel splits to generate data that is subsequently used to train a combustion model. The trained model may then be used in an optimization system to automatically calculate the appropriate fuel split for tuning combustion performance. According to alternative embodiments, additional inputs and outputs may be included in the optimization, such as IGV settings, bypass gas, flashback, etc. As will be seen, several performance advantages may be achieved, including those related to improved gas turbine emissions and burner dynamics. As an additional advantage, the present disclosure may provide gas turbine operators with the flexibility to manage operational tradeoffs in real time. For example, an operator may prioritize emissions or combustor dynamics, or some weighted combination thereof.

[0057] Reference now Figure 6 , an exemplary combustion auto-tuning controller or system (also referred to as a "combustion auto-tuner") 130 is shown according to the present disclosure. The combustion auto-tuner 130 can be a stand-alone computer-implemented controller or control system, for example, which is integrated within the GT controller, for example, relative to Figure 2 In addition, the combustion auto-tuner 130 may be distributed in Figure 2 The combustion auto-tuner 130 controls or tunes the combustion system of the gas turbine to maintain proper performance under variable conditions such as ambient temperature, humidity, and pressure, all of which change over time and have a significant impact on combustion operations. The combustion auto-tuner 130 can also maintain tuning of the gas turbine as components wear or degrade and during changes in fuel composition. It should be understood that changes in fuel composition may cause changes in heat release, which may result in unacceptable emissions, unstable combustion, or even LBO. As will be seen, the combustion auto-tuner 130 of the present disclosure does not rely on adjusting the fuel composition to compensate for these variables. Instead, the combustion auto-tuner of the present disclosure tunes operating parameters (e.g., fuel split, fuel temperature, turbine fuel / air ratio, etc.) to bring combustion performance back within the desired operating parameters.

[0058] like Figure 6As shown, the controlled variables (or "CVs") of the combustion auto tuner 130 may include NOx, CO, and combustor dynamics. Disturbance variables (or "DVs") may include load percentage and compressor inlet temperature. Other disturbance variables (not shown) such as ambient humidity and pressure may also be included. According to an exemplary embodiment, the manipulated variables (or "MVs") include a first fuel split bias and a second fuel split bias, as well as settings related to bypass air and IGV. According to an alternative embodiment, other MVs (not shown) may include a fuel temperature setting and a total fuel / air ratio.

[0059] As used herein, "fuel split" refers to a setting that determines the distribution of fuel supply between various nozzles or nozzle types within a combustor. For example, the fuel split of a combustor may determine how the combustor fuel supply is distributed between the main nozzles, pilot nozzles, and / or late lean nozzles or downstream injection nozzles. Figure 6 As shown in the example of , MV can be described by the "fuel split bias" of different nozzles of the burner. As used herein, the fuel split "bias" parameter represents the percentage change in the fuel split that would otherwise be applied to a particular nozzle. As described above, the different nozzles of the burner may generally be referred to as the first nozzle, the second nozzle, and the third nozzle (if present). In this regard, the "first fuel split bias" refers to the fuel split bias associated with the first nozzle, the "second fuel split bias" refers to the fuel split bias associated with the second nozzle, and, when the third nozzle is present, the "third fuel split bias" refers to the fuel split bias associated with the third nozzle. Thus, for example, a first fuel split bias of negative 1 (-1) indicates that the fuel split of the first nozzle will be reduced by one percent (1%). Or, for example, if the fuel split of the second nozzle is 12%, and the combustion auto tuner 130 calculates a value of positive 2 (+2) for the second fuel split bias, the fuel split of the second nozzle will be adjusted to 14%.

[0060] According to the system described herein, data received from various sensors related to combustor dynamics, exhaust emissions, and other relevant combustion parameters may be transmitted to a GT controller, such as a fleet controller 39 and / or a combustion auto tuner 130. Figure 6As shown, the combustion auto-tuner 130 may receive target operating parameters 132 and may compare received sensor data to the target operating parameters to determine whether the combustion system is operating in accordance therewith. According to a preferred embodiment, these target operating parameters 132 may be based on a priority setting 134, which is defined at least in part based on input received from an operator. If the combustion system of the gas turbine is operating within the target operating parameters, the setting of the MV may be maintained unchanged by the GT controller. However, if the combustion system of the gas turbine is not operating within the target operating parameters, the combustion auto-tuner 130 may recommend a series of adjustments to the MV to achieve the target operation, which may be implemented by the GT controller.

[0061] As will be described in more detail below, the combustion auto-tuner 130 may include an optimization system 136, and the recommended adjustments to the MV may be based on its operation. The optimization system 136 may include a model 137 and an optimizer 138 that model the combustion system of the gas turbine. The optimization system 136 may use the model 137 and the optimizer 138 to optimize a cost function that may correspond to a priority setting received from an operator. An interface display may be provided that allows an operator to select between different operational priorities for the combustion system and / or the gas turbine. For example, such a priority setting may include alternatives that prioritize emissions and combustor dynamics differently, and the priority may be reflected in the cost function used in the optimization. Operational priorities may include: minimizing NOx, CO, or combustor dynamics; or maximizing generated load, efficiency, etc. Such functionality may allow an operator to switch between operational priorities in response to changing market conditions, gas turbine performance characteristics, environmental conditions, etc.

[0062] Reference now Figure 7 , shows a schematic diagram of a method 140 for monitoring and collecting data related to combustor dynamics according to an embodiment of the present disclosure. At box 141, for example, the previously introduced CDMS 115 may measure the vibration response or frequency generated by each combustor of the gas turbine during operation. As shown in the second box 142, the measured frequency of each combustor of the gas turbine may be normalized by a derived normalization constant. In the next box 143, a maximum value may be determined and output to the combustion auto-tuner 130 as a combustor maximum normalized pressure (or "CMNP"). According to the present disclosure, CMNP represents a specific type of combustor dynamic data that the combustion auto-tuner 130 can use for effective combustion tuning.

[0063] Reference now Figure 8, shows an alternative combustion auto-tuner 130, which includes the CMNP and NOx introduced above as controlled variables. The combustion auto-tuner 130 can be used to keep both NOx and CMNP below corresponding predetermined limits. To this end, the combustion auto-tuner 130 can use the fuel split bias as the MV to tune the combustion, which can include determining the first fuel split bias, the second fuel split bias, and the third fuel split bias corresponding to the first nozzle, the second nozzle, and the third nozzle, respectively. According to an exemplary embodiment, the first nozzle may represent a pilot nozzle, the second nozzle may represent a main nozzle, and the third nozzle may represent a downstream injection nozzle. Other configurations involving other nozzle types are also possible.

[0064] The combustion auto-tuner 130 may include a limit that the net change in fuel split bias is equal to zero (0). For example, such a limit may take the form of:

[0065] [1st fuel split bias] - [2nd fuel split bias] - [3rd fuel split bias] = 0 This type of condition restricts the problem to two free dimensions. Since load percentage and inlet temperature also affect the controlled variables, namely, NOx and CMNP, these are also included as DVs.

[0066] refer to Fig. 9 , illustrates an exemplary operation of the optimization system 136 of the combustion auto-tuner 130. As described above, the optimization system 136 may include an optimizer 138 and a model 137. Fig. 9 In the embodiment shown in FIG. 1 , the optimization system 136 can be part of a combustion auto-tuner 130 that communicates with a unit controller 39. The unit controller 39 can be a computer-based controller that controls the operation of the gas turbine 27, which can be located, for example, within the power plant 25. The combustion auto-tuner 130 can be a computer-implemented controller that provides supervisory or advisory control to the unit controller 39. It should be understood that in alternative embodiments, Fig. 9 The various computer-implemented components shown in the drawings may reside on the same computer or may be distributed across multiple different computer systems. For example, the model 137 may reside on a different computer system than the optimizer 138. In addition, an operator interface (not shown) may provide a means for an operator to communicate with the fleet controller 39 and / or the combustion auto-tuner 130. Figure 2 As depicted, the gas turbine 27 may include a plurality of actuators 57 and sensors 56. The sensors 56 may include devices for sensing and measuring various operating parameters, while the actuators 57 may include devices for actuating components such as valves and dampers to effectuate control.

[0067] It should be understood that the following will refer to Figures 10 to 16The model 137, discussed in more detail, represents the relationship between (a) manipulated variables and disturbance variables and (b) controlled variables of the combustion system of the gas turbine. Manipulated variables or MVs represent those variables that can be changed by an operator or the optimization system 136 to affect the controlled variables or CVs, and may include any of those listed. Disturbance variables or DVs refer to those variables that affect the controlled variables but cannot be manipulated by the operator, and may include any of those listed. It should be understood that the optimizer 138 can function by determining a set of enhanced or optimal set values ​​for the MVs based on (1) desired objectives associated with the operation of the gas turbine 27 (e.g., minimizing NOx, CO, or combustor dynamics) and (2) constraints associated with the operation of the gas turbine 27 (e.g., emissions limits).

[0068] At a predetermined frequency, the optimization system 136 may obtain the current values ​​of the manipulated variables, controlled variables, and disturbance variables from the unit controller 39. Each time the current values ​​of the manipulated variables, controlled variables, and disturbance variables are read from the unit controller 39, an "optimization cycle" begins. The optimization system 136 may then use the model 137 and the optimizer 138 to determine a set of optimal set values ​​for the manipulated variables based on the current conditions of the gas turbine 27. The combustion auto-tuner 130 may then send this set of optimal set values ​​to the unit controller 39. The unit controller 39 may be configured to automatically implement this set of optimal set values ​​in the control of the gas turbine 27. Alternatively, the operator of the power plant 25 may have the option of approving the unit controller 39 to use this set of optimal set values. The combustion auto-tuner 130 and the optimization system 136 may operate in a closed loop to adjust the set values ​​of the manipulated variables at a predetermined frequency of the optimization cycle (e.g., as frequently as once every second or every 10 seconds, or as little as once every half an hour).

[0069] In deriving this set of optimal set values, the optimizer 138 can use the model 137 to predict the future operation of the combustion system in order to minimize a cost function subject to a set of constraints. The cost function is a mathematical representation of the desired operating objectives of the gas turbine. For example, to minimize NOx, the cost function may include terms that decrease as the NOx level decreases. A common method for minimizing the cost function is called "gradient descent optimization". Gradient descent is an optimization algorithm that approaches the local minimum of a function by taking a step size proportional to the negative value of the gradient (or approximate gradient) of the function at the current point. Constraints may be imposed on both the input (MV) and the output (CV) of the gas turbine 27 at future times. Typically, constraints consistent with the restrictions associated with the unit controller 39 are imposed on the manipulated variables. The constraints on the output (CV) are determined by the problem being solved.

[0070] According to a preferred embodiment, optimizer 138 uses nonlinear programming ("NLP") optimization to solve the optimization problem. However, it should be understood that a variety of different optimization techniques can also be used according to the form of the model, cost function and constraints. For example, it is expected that optimizer 138 can be implemented by using a variety of different types of optimization methods alone or in combination. These optimization methods include but are not limited to linear programming, quadratic programming, mixed integer nonlinear programming, stochastic programming, global nonlinear programming, genetic algorithms and particle swarm technology. NLP optimizer can solve the problem of 20 manipulated variables and 10 controlled variables in less than one second. For the present application, this is fast enough because the optimization cycle is usually in any range from less than 1 second to 20 minutes.

[0071] The model 137 of the optimization system 136 is configured to capture the relationship between the manipulated variables and the disturbance variables and the controlled variables of the combustion system of the gas turbine. According to an exemplary embodiment, the model 137 used by the optimization system 136 can take several different forms. For example, such a model can be developed based on: 1) known first principle equations describing the system, resulting in a physics-based model; 2) data, resulting in an empirical model; or 3) a combination of first principle equations and data. However, when modeling NOx and CO emissions and combustor dynamics in gas turbine combustion, it is difficult to derive and maintain a model based on first principle equations. To this end, according to a preferred embodiment, the model 137 includes a data-driven or empirical model that is based on data collected from actual operation of the gas turbine and its combustion system.

[0072] According to a preferred embodiment, model 137 is a nonlinear model. In order to correctly capture the relationship between the manipulated / disturbance variables and the controlled variables associated with gas turbine combustion, a nonlinear model may be required. A nonlinear model may represent a curvilinear relationship rather than a straight-line relationship, which is common for complex systems such as those discussed herein. In view of the foregoing requirements, a neural network-based approach is currently a preferred embodiment of an empirical model for implementing model 137. A neural network is developed based on empirical data using an advanced regression algorithm. See, for example, C. Bishop's "Pattern Recognition and Machine Learning", Springer, New York, New York, 2006, which is incorporated herein by reference in its entirety. Neural networks can effectively capture the nonlinearities that complex systems such as gas turbines typically exhibit. Therefore, according to an exemplary embodiment, model 137 of the optimization system 136 is a data-driven or empirical model, that is, it is based on collected data. In addition, according to a preferred embodiment, the type of empirical model used for model 137 is a neural network. In addition, other types of empirical modeling techniques can be used to model the combustion system. For example, model 137 may also include a type of model selected from the following list: support vector machine model, random forest model, decision tree model, k-means model, principal component analysis ("PCA") model, gradient boosting model and parameter equation fitting model.

[0073] Reference now Fig.10 , Fig.11 , Fig.12 and Fig.13 The optimization system 136 of the combustion auto-tuner 130 may use the model 137 according to several alternative methods. For example, Fig.10 As shown, model 137 is configured to predict future values ​​of CVs (such as NOx, CO, and / or combustor dynamics (i.e., "CMNP")) using disturbance rejection, as further described in U.S. Pat. No. 7,123,971, which is incorporated herein by reference in its entirety. As shown, model 137 is used to predict CVs at time (t) using MV and DV at time (t), which are then compared to the actual or measured values ​​of CVs at time (t). The difference between the prediction of the controlled variable at time (t) and the actual value of the controlled variable at time (t) is then calculated and added to the prediction of NOx by model 137 at time (t+1). In this way, the prediction at time (t+1) is biased by the error in the prediction at time (t).

[0074] Disturbance rejection methods have proven to be useful for optimization processes. This is because, in optimization, the interest is in minimizing, for example, NOx at time (t+1) by determining a set of values ​​for the MV at time (t). The error in the model at the current time is known and can be used to bias the prediction at time (t+1). In this way, disturbance rejection allows slowly varying unmeasured disturbances to be suppressed in the model predictions. To this end, Fig.10 The method is a preferred embodiment.

[0075] refer to Fig.11 , provides an alternative model configuration in which the sampling model 137 has a fixed bias set added to it. It should be understood that the naming on the output indicates the mode of the fixed bias set. Fig.12 , shows a simplified version of this type of model. In this case, the fuel split bias is summarized as MV, while the load percentage and inlet temperature are summarized as DV. In addition, since this model will be used to optimize the MV in the future, that is, when the value of DV is unknown, it is assumed that DV does not change, so: DVs(t) = DVs(t+1). As shown in the figure, the error is fixed and calculated, where the mode (0,0,1) indicates how to add a fixed bias set.

[0076] refer to Fig.13 , shows an aggregate model for predicting CVs such as NOx, CO and / or combustor dynamics (i.e., "CMNP"). The aggregate model can be used to provide a weighted average of CVs around a MV. According to an exemplary embodiment, this form can be used to penalize bias values ​​in the optimization, which produces a high value or limit close to the setting for the CV. In this case, if the value is within a predetermined distance of these limits, the value is penalized. In this way, the optimization can be used to optimize the continuously updated value of the MV. For example, a cost function can be constructed, which is a function of the MV. Constraints can be included, which are also functions of the MV. Although the cost function and constraints must be continuous functions, they can be nonlinear. An upper limit or upper bound can also be set on the MV. Then, given the MV, cost function, constraints, and MV limits, a nonlinear programming ("NLP") solver can be used to find the MV subject to the limits and constraints and minimize the cost function.

[0077] More specifically, for example, such a cost function may take the following form:

[0078] Cost=(NOx(t+1) / s 1 ) 2 +(CMNP(t+1) / s 2 ) 2

[0079] Among them, NOx(t+1) and CMNP(t+1) are predicted by the aggregation model, and s1 and 2 is a scaling factor for weighing NOx and CMNP performance. Constraints may include, for example, that the net change in fuel split bias is equal to zero (0). And, limits may include, for example:

[0080] F+>First fuel split bias (t+1)>F-

[0081] S+>Second fuel split bias (t+1)>S-

[0082] T+>Third fuel split bias (t+1)>T-

[0083] Wherein, the values ​​F+, F-, S+, S-, T+ and T- are fixed values ​​predetermined before the optimization run. Given the cost, constraints and bounds, an optimizer such as an NLP optimizer can determine the MV(t+1) that minimizes the cost (i.e., C subject to the defined constraints and bounds).

[0084] The optimizer 138 may run periodically, for example, every 300 seconds, 27 seconds, 10 seconds, 5 seconds, or 1 second, and may calculate an optimal MV, which may then be used to bias the fuel split and / or other MVs of the gas turbine. Thus, the systems and methods of the present disclosure may optimize, tune, and / or control various aspects of combustion and combustor performance within a gas turbine.

[0085] The present disclosure will now discuss other aspects of exemplary empirical models such as neural networks, and issues associated with "experimental design" for collecting data and then training models based on the collected data. Since the settings of the manipulated variables for a gas turbine may not be substantially changed during normal operation, it is often necessary to perform "experimental design" to collect the operational data required to build and train such data-driven models.

[0086] Fig.14 The results of an exemplary "experimental design" for collecting data for training an empirical model such as a neural network are shown. As shown, time series plots of two MVs, one DV, and one CV of interest are provided, where the CV of interest is depicted as NOx emissions. MV 1 For example, it may represent the first fuel split offset associated with the first nozzle, while MV 2For example, it may represent a second fuel split bias associated with a second nozzle. DV may represent, for example, a load percentage or a compressor inlet temperature. It should be understood that although only 2 MVs, 1 DV, and 1 CV are shown in this example, when a data collection run is performed according to an experimental design, data is typically collected for multiple MVs, multiple DVs, and multiple CVs. The sampling time of the data is provided at the top of the figure. In this case, sampling starts at 11:00 a.m. on day 1 and ends at 11:00 a.m. on day 2, with samples collected every 15 minutes. The first sample is marked as time t=0, the second sample is marked as time t=1, and so on, until the last time point t=T.

[0087] It should be understood that the goal of the experimental design is to collect data sufficient to develop or train an empirical model so that the model can then be used to optimize combustion performance. To perform a proper experimental design, the MVs can be moved independently so that the movements are uncorrelated over time. Fig.14 As shown, at each sampling time, the MVs are moved (if they are to be moved within the sampling period). After this, the MVs are held constant until the next sampling time. In the illustrated case, the interest is in establishing a steady-state model, and therefore, the sampling time is long enough to allow the process to reach a steady state. Therefore, since the movement of the MVs may take several minutes or longer to fully realize its effect, the sampling time can be adapted to this lag. (In this application, the focus is primarily on steady-state models. However, it should be understood that the systems and methods of the present invention can be extended to dynamic models.) For example, as Fig.14 As shown, mobile MV 1 And the effect of keeping it stable from time (t=1) to just before time (t=2) affects the NOx level at time (t=2).

[0088] Fig.14 The MV, DV and NOx sampling points depicted in can be used as a training dataset to train empirical models, e.g. Figure 6 and Figure 8 Model 137. In the case where the model includes a neural network and / or interference suppression, the model can be trained via the methods and systems described in detail in U.S. Patent 7,123,971 and U.S. Patent Application Publication No. 20180024512, both of which are incorporated herein by reference in their entirety. Once the model is trained, the model can be used in any optimization system 136 described herein.

[0089] Fig.15A method 200 is shown that involves an automated process for collecting data to train an empirical model, such as a neural network, that can be used as a model 137 within an optimization system 136 in accordance with the present disclosure. The method 200 may begin by determining whether the model requires additional data or "learning" (block 202). If it is determined that learning is required, and it is determined that the load of the gas turbine is stable enough for data collection, a "learning mode" may be enabled (block 204) and a "learning run" may be continued while the gas turbine is operating for a period of time (e.g., 1 hour). The learning run may be configured as described above in Fig.14 The collected data may then be used to further train the model 137. If it is determined that additional learning is not required, the learning mode may be deactivated or remain deactivated (block 206).

[0090] As further shown, method 200 may include a maximum idle time and a minimum idle time that control aspects of the enabling of learning mode. For example, a value for the maximum idle time (box 208) may be set, such as 5 days, after which the learning mode is automatically enabled even if learning is not determined to be required. In this way, the maximum idle time ensures that the time between learning runs will never exceed a predetermined maximum time period. A value for the minimum idle time (box 209) may also be set, such as 1 day. In this way, for example, once the learning mode is turned on and data is collected, there is a guaranteed minimum time period before performing another learning run. However, if the time since the last learning run is between the maximum idle time and the minimum idle time, it is the determination of "is learning required?" (box 202) that controls whether the learning mode is enabled. A delay (box 210) can be used to limit the frequency with which method 200 loops through the analysis.

[0091] According to a preferred embodiment, the "Is learning required?" determination (block 202) is performed in the following manner. First, a historical training data database of previous learning runs, moves, and collected training data within a predefined previous time period (e.g., the last 6 or 12 months) is queried. From this query, the data requirement can be determined. Using the historical training data database, the Euclidean distance from the current operating point (e.g., defined by load percentage and inlet temperature) to the five nearest neighboring data points in the historical training data database is calculated. If the average distance to the five nearest neighboring data points is greater than a predefined threshold, it is determined that insufficient data is available at the current load and temperature, and therefore learning is determined to be required. Other methods for making this determination are also possible.

[0092] Fig.16An exemplary learning run 220 performed during a learning mode is shown for collecting or augmenting a training data set for training an empirical model according to the present disclosure. In general, it should be understood that a learning run is performed to more fully populate the training data set with data representing operation under a wide range of potential operating conditions. A more complete training data set produces a better model.

[0093] The learning run 220 may begin at the current operating condition defined as (0,0), i.e., the first fuel split bias is zero and the second fuel split bias is zero. In the exemplary case shown, only two dimensions are required, however, it should be understood that three dimensions may be used to include another fuel split bias. The learning run 220 may include several incremental steps in which only one of the first fuel split bias and the second fuel split bias is moved at a time. Thus, for example, step 1 may include a move in which the second fuel split bias is moved from 0.0 to 0.5%. The fuel split bias may then be allowed to stabilize for a period of time (e.g., 5 minutes or 10 minutes or 15 minutes) so that the measured value of the CV (e.g., emission levels or combustion dynamics) includes any change in value caused by the step 1 move. Step 2 may then be performed in which the fuel split bias is moved from (.5,0) to (1.0,0). Again, the fuel split bias may be allowed to stabilize for a period of time and the steady-state measurement result recorded. The example may proceed with steps 3, 4, and 5 as shown. After performing the step 5 move, high combustion dynamics are observed, including, for example, dynamics that register as being above a predetermined threshold. Since high dynamics may be damaging to the machine, the next step, step 6, may occur immediately upon the high dynamics being observed. Thus, the machine may only be at (1.0, 1.5) for a few seconds, rather than the longer duration of the other steps. Steps 7, 8, and 9 may then be performed, with a hold period and measurements of steady state operation recorded. In this way, data reflecting operation under a range of operating conditions may be collected in a single learning run.

[0094] According to an exemplary embodiment, a learning run may be configured to collect new data relating to untested combinations of MVs and DVs and / or combinations of MVs and DVs that have not been recently tested, collectively referred to as "underrepresented data" within a training data set. That is, embodiments of the present disclosure may emphasize the collection of underrepresented data in learning runs in which they are enabled. To this end, the system of the present disclosure may monitor operating conditions occurring during gas turbine operation and determine those situations where current operating conditions are equal to or approximately equal to operating conditions that were found to be underrepresented in the training data set. This functionality may enable the most needed data in the training data set to be collected more efficiently and economically. For example, the present embodiment may monitor gas turbine operation, compare current conditions to a list of priority or underrepresented data, and determine when an opportunity arises to collect that data without significantly deviating from the current operating plan of the gas turbine.

[0095] Thus, for example, an exemplary mode of operation of the present embodiment may include performing a model training process including the following. Additional data may be acquired for inclusion in an original training data set in order to create a modified training data set. The additional data may include measurements of inputs and outputs of sensors during a selected period of operation of the combustion system. A model of the optimization system may then be trained based on the modified training data set. The step of acquiring additional data may include an automatic learning mode including: analyzing the original training data set to determine at least one data requirement; generating an experimental design for acquiring the data requirement; monitoring the operation of the gas turbine to determine an opportunity to collect the data requirement based on the generated experimental design; and transmitting a prompt to the gas turbine when an opportunity to collect the data requirement is determined. For example, the prompt may describe a change to one of the manipulated variables based on the experimental design, and the control system may then implement the change to collect the data requirement and add the data requirement to the data set used to train the model.

[0096] As another example, according to the system described herein, a historical training data database may be included, which stores records describing all training data in an original training data database. The step of analyzing the original training data set to determine data needs (i.e., underrepresented data) may include determining a combination of manipulated variables and interference variables that: is not included in the historical training data database; or has not been updated in the historical training data database within a predetermined time limit. The step of monitoring the operation of the gas turbine to determine an opportunity to collect data needs may include: determining a combination of manipulated variables and interference variables associated with current conditions of operation of the gas turbine; and comparing the combination of manipulated variables and interference variables associated with the current conditions of operation of the gas turbine with the combination of manipulated variables and interference variables associated with the data needs. When close enough, an opportunity to collect data needs is determined.

[0097] As will be appreciated by those of ordinary skill in the art, the features and configurations of many variations described above with respect to several exemplary embodiments may be further selectively applied to form other possible embodiments of the present invention. For the sake of brevity and in view of the capabilities of those of ordinary skill in the art, all possible iterations are not provided or discussed in detail, but all combinations and possible embodiments are covered in the following claims, or are otherwise intended to be part of the instant application. In addition, based on the above description of several exemplary embodiments of the present invention, those skilled in the art will recognize the relevant improvements, changes and modifications. These improvements, changes and modifications within the technical scope of the art are also intended to be covered by the appended claims. In addition, it should be apparent that the foregoing only relates to the embodiments of the present application, and many changes and modifications may be made herein without departing from the essence and scope of the present application as defined by the following claims and their equivalents.

Claims

1. An electric power system (10), comprising: a gas turbine (27) having a combustion system; a control system (39) operatively connected to the gas turbine (27) for controlling operation thereof; as well as A combustion auto-tuner (130) communicatively coupled to the control system (39), the combustion auto-tuner (130) comprising: An optimization system (136), the optimization system comprising: a model (137), the model comprising an empirical model, the empirical model models the combustion system by statistically mapping inputs to outputs of operations of the combustion system, thereby generating a predicted value of the output at a future time based on the inputs of the combustion system, wherein the model (137) comprises a class of models selected from the following list: a neural network, a support vector machine model, a random forest model, a decision tree model, a k-means model, a principal component analysis model, a gradient boosting model, and a parametric equation fitting model; and Optimizer(138); a sensor (56) configured to measure the input and the output of the operation of the combustion system, the output of the combustion system comprising at least one controlled variable and the input of the combustion system comprising a disturbance variable and a manipulated variable; wherein the model (137) includes a representation of a relationship between (a) the manipulated variable and the disturbance variable and (b) the at least one controlled variable of the combustion system; and wherein the optimized control scheme for the combustion system includes a set value for at least one of the manipulated variables; Hardware processor; and a machine-readable storage medium storing instructions thereon, the instructions causing the hardware processor to perform a tuning process for tuning the operation of the combustion system, wherein the machine-readable storage medium further comprises instructions causing the hardware processor to perform a model training process; The model training process includes: acquiring additional data for inclusion in an original training data set to create a modified training data set, the additional data comprising measurements of the input and the output of the sensor during a selected period of operation of the combustion system; and training the model of the optimization system based on the modified training data set; The tuning process includes the following steps: receiving current measurements of the input and the output from the sensor (56); using the optimization system (136) to calculate an optimized control solution for the combustion system based on the current measurements received from the sensor (56); and transmitting the optimized control scheme to the control system (39), and Wherein, the step of acquiring the additional data includes an automatic learning mode, and the automatic learning mode includes: analyzing the original training data set to determine at least one data requirement; generating an experimental design for acquiring the at least one data requirement; monitoring said operation of said gas turbine for determining opportunities to collect said at least one data requirement according to said generated design of experiments; and transmitting a prompt to the gas turbine (27) when the opportunity to collect the at least one data requirement is determined; The prompt describes a change in a first manipulated variable among the manipulated variables according to the experimental design.

2. The system (10) according to claim 1, wherein: The model (137) comprises a neural network model; and wherein, to calculate the optimal control scheme, the optimizer (138) uses the model to predict future operation of the combustion system so as to minimize a cost function subject to a set of constraints.

3. The system (10) of claim 1, wherein: The machine-readable storage medium further comprises a historical training data database, wherein the historical training data database stores records describing the training data in the original training data set; Wherein, the step of analyzing the original training data set to determine the at least one data requirement comprises determining a combination of the manipulated variable and the disturbance variable, the combination: is not included in the historical training data database; or has not been updated in the historical training data database within a predetermined time limit.

4. The system (10) according to claim 3, wherein: Upon receipt, the control system automatically controls the gas turbine (27) in accordance with the change to the first of the manipulated variables described by the prompt; and wherein the step of monitoring the operation of the gas turbine (27) to determine an opportunity to collect the at least one data requirement comprises: determining a combination of the manipulated variable and the disturbance variable associated with a current condition of the operation of the gas turbine; The combination of the manipulated variable and the disturbance variable associated with the current condition of the operation of the gas turbine is compared to the combination of the manipulated variable and the disturbance variable associated with the at least one data requirement.

5. The system (10) of claim 1, wherein: The optimization system (136) includes a cost function including a mathematical representation for evaluating future operation of the combustion system relative to one or more operational priorities and one or more operational constraints of the combustion system; and The tuning process further includes: the optimizer determining the set value of at least one of the manipulated variables by accessing the model to minimize the cost function.

6. The system (10) according to claim 5, wherein: The optimizer (138) of the optimization system (136) is selected from the group consisting of linear programming, quadratic programming, mixed integer nonlinear programming, gradient descent optimization, stochastic programming, global nonlinear programming, genetic algorithms, and particle / swarm techniques; wherein the control system automatically controls the gas turbine according to the set value of the at least one of the manipulated variables; and The combustion auto-tuner (130) and the optimization system (136) operate in a closed loop to adjust the set value of at least one of the manipulated variables at a predetermined frequency according to an optimization period.

Images