Control and tuning of gas turbine combustion

Patent metadata

Field	Value
Publication number	US-10626817-B1
Application number	US-201816144548-A
Country	US
Kind code	B1
Filing date	Sep 27, 2018
Priority date	Sep 27, 2018
Publication date	Apr 21, 2020
Grant date	Apr 21, 2020

Abstract

A system that includes: a gas turbine having a combustion system; a control system operably connected to the gas turbine for controlling an operation thereof; and a combustion auto-tuner, which is communicatively linked to the control system, that includes an optimization system having an empirical model of the combustion system and an optimizer; sensors configured to measure the inputs and outputs of the combustion system; a hardware processor; and machine-readable storage medium on which is stored instructions that cause the hardware processor to execute a tuning process for tuning the operation of the combustion system. The tuning process includes the steps of: receiving current measurements from the sensors for the inputs and outputs; given the current measurements received from the sensors, using the optimization system to calculate an optimized control solution for the combustion system; and communicating the optimized control solution to the control system.

First claim

That which is claimed:

1. A system comprising:
a gas turbine having a combustion system;
a control system operably connected to the gas turbine for controlling an operation thereof; and
a combustion auto-tuner communicatively linked to the control system, the combustion auto-tuner comprising:
an optimization system comprising:
a model, the model comprising an empirical model that models the combustion system via statistically mapping inputs to outputs of an operation of the combustion system so to generate predicted values for the outputs at future times based on the inputs of the combustion system, the model comprising a type of model selected from the following list: a neural network, a support vector machine model, a random forest model, a decision trees model, a k-means model, a principal component analysis model, a gradient boost model, and a parametric equation fitting model; and
an optimizer;

sensors configured to measure the inputs and outputs of the operation of the combustion system, the outputs of the combustion system comprising at least one controlled variable, and the inputs of the combustion system comprising disturbance variables and manipulated variables;
wherein the model comprises a representation of a relationship between (a) the manipulated variables and the disturbance variables and (b) the at least one controlled variable of the combustion system; and
wherein the optimized control solution for the combustion system comprises a setpoint value for at least one of the manipulated variable;
a hardware processor; and
a machine-readable storage medium on which is stored instructions that cause the hardware processor to execute a tuning process for tuning the operation of the combustion system, wherein the machine-readable storage medium comprises instructions that cause the hardware processor to execute a model training process;
wherein the model training process comprises:
obtaining additional data for inclusion within an original training dataset in order to create a revised training dataset, the additional data comprising measured values of the inputs and outputs by the sensors during a selected period of operation of the combustion system; and
training the model of the optimization system pursuant to the revised training dataset;
wherein the tuning process comprises the steps of:
receiving current measurements from the sensors for the inputs and outputs;
given the current measurements received from the sensors, using the optimization system to calculate an optimized control solution for the combustion system; and
communicating the optimized control solution to the control system; and
wherein the step of obtaining the additional data comprises an automated learning mode that includes:
analyzing the original training dataset to determine at least one data need;
generating a design of experiment for acquiring the at least one data need;
monitoring the operation of the gas turbine for determining an opportunity to gather the at least one data need pursuant to the generated design of experiment; and
communicating a prompt to the gas turbine when the opportunity to gather the at least one data need is determined;
wherein the prompt describes a change to a first one of the manipulated variables in accordance with the design of experiment.

2. The system of claim 1 , wherein the model comprises a neural network model; and
wherein, to calculate the optimized control solution, the optimizer uses the model to predict a future operation of the combustion system in order to minimize a cost function subject to a set of constraints.

3. The system of claim 1 , wherein the machine-readable storage medium further comprises a historical training data database in which is stored records describing training data in the original training database;
wherein the step of analyzing the original training dataset to determine the at least one data need comprises determining combinations of the manipulated and disturbance variables that:
are not included within the historical training data database; or
have not been updated in the historical training data database within a predetermined time limit.

4. The system of claim 3 , wherein, upon receiving, the control system automatically controls the gas turbine pursuant to the change to the first one of the manipulated variables described by the prompt; and
wherein the step of monitoring the operation of the gas turbine for determining an opportunity to gather the at least one data need comprises:
determining a combination of the manipulated and disturbance variables associated with current conditions of the operation of the gas turbine;
comparing the combination of the manipulated and disturbance variables associated with the current conditions of the operation of the gas turbine against the combination of the manipulated and disturbance variables associated with the at least one data need.

5. The system of claim 1 , wherein the optimization system includes a cost function that comprises a mathematical representation for evaluating the future operation of the combustion system relative to one or more operating priorities and the one or more operating constraints of the combustion system; and
wherein the tuning process further comprises the optimizer determining the setpoint value for the at least one of the manipulated variables by accessing the model to minimize the cost function.

6. The system of claim 5 , wherein the optimizer of the optimization system is selected from a group consisting of: linear programming, quadratic programming, mixed integer non-linear programming, gradient descent optimization, stochastic programming, global non-linear programming, genetic algorithms, and particle/swarm techniques;
wherein the control system automatically controls the gas turbine pursuant to the setpoint value for the at least one of the manipulated variable; and
wherein the combustion auto-tuner and the optimization system run in a closed loop adjusting the setpoint value of the at least one of the manipulated variables at a predetermined frequency in accordance with an optimization cycle.

7. The system of claim 5 , wherein the combustion system comprises a combustor having at least two type of fuel nozzles: a first nozzle and a second nozzle; and
wherein the disturbance variables include at least one of a load percentage and a compressor inlet temperature.

8. The system of claim 7 , wherein the at least one of the manipulated variables includes a fuel split to the combustor that describes how a fuel supply is divided between the first nozzle and the second nozzle.

9. The system of claim 8 , wherein the controlled variables comprise at least one of a level of NOx emissions and a level of CO emissions for the gas turbine.

10. The system of claim 8 , wherein the controlled variables comprise combustor dynamics.

11. The system of claim 8 , wherein the model comprises disturbance rejection;
wherein the disturbance rejection comprises a configuration of the model in which the predicted value made by the disturbance rejection model for the output at the future time is based upon:
a predicted value made by the model for the output at the future time; and
a value of a bias that is based upon an error that the difference between previous corresponding measured and predicted values.

12. The system of claim 8 , wherein the empirical model comprises a neural network that includes multiple layers having nodes, the multiple layers including at least an input layer, an output layer, one or more hidden layers, and forward weight matrixes; and
wherein:
the input layer comprises a plurality of the nodes,

Assignees

• Gen Electric

Inventors

• Piche Stephen William
• Pickard Fred Francis
• Petro Robert Nicholas
• Liu Yan
• Virani Nurali

Classifications

• G06N3/08

  Learning methods · CPC title

• F02D41/1461

  of the exhaust gases emitted by the engine · CPC title

• F23N2900/05006

  Controlling systems using neuronal networks · CPC title

• F23N2223/48

  Learning / Adaptive control · CPC title

• F23N2225/00

  Measuring · CPC title