| CPC F01K 23/101 (2013.01) [F01K 13/02 (2013.01); F05D 2220/722 (2013.01); F05D 2260/2322 (2013.01); F05D 2260/80 (2013.01); F05D 2260/85 (2013.01); F05D 2270/053 (2013.01); F05D 2270/709 (2013.01)] | 14 Claims |
|
1. A processor implemented method for optimizing an operation of a combined cycle gas turbine (CCGT) plant, the method comprising:
receiving a plurality of data from a one or more databases of the CCGT plant at a predetermined frequency, wherein the plurality of data comprises of a real-time and a non-real-time data, wherein the real-time data is obtained from plant automation systems via a communication server;
preprocessing, via one or more hardware processors, the plurality of data;
estimating, via the one or more hardware processors, a set of soft sensor parameters using a plurality of soft sensors;
integrating, via the one or more hardware processors, the set of soft sensor parameters with the pre-processed plurality of data, wherein the integrated data comprises a first set of manipulated variables;
training, via the one or more hardware processors, a plurality of anomaly detection models and a plurality of anomaly diagnosis models using a historical data of the CCGT plant and built using statistical, machine learning and deep learning techniques including principal component analysis, Mahalanobis distance, isolation forest, random forest classifiers, one-class support vector machine, artificial neural networks, elliptic envelope and auto-encoders and Bayesian networks;
detecting, via the one or more hardware processors, process and equipment anomalies related to the CCGT plant and individual units of the CCGT plant, using the plurality of anomaly detection models, wherein the plurality of anomaly detection models are retrieved from the database;
identifying, via the one or more hardware processors, at least one cause of the detected anomalies using the plurality of anomaly diagnosis models, wherein the plurality of anomaly diagnosis models are retrieved from the database, and wherein a real-time process optimization is triggered only in an absence of the anomaly;
determining, via the one or more hardware processors, a state of the operation of the CCGT plant using a plurality of state determination models wherein the state can be steady or unsteady state;
classifying, via the one or more hardware processors, the state of the operation of the CCGT plant into the steady state and the unsteady state in real-time using a subset of the CCGT plant variables comprising a total generated power, a frequency of power generated or a rotational speed of shaft, a fuel flow rate and an inlet air flow rate using a plurality of state determination which are data-driven classifiers, wherein the unsteady state is further classified into one of a steady, load-up, load-down, start-up and shut-down state;
predicting, via the one or more hardware processors, a plurality of key performance parameters of CCGT plant using a plurality of predictive models and the integrated data, wherein the plurality of predictive models are retrieved from the database;
configuring, via the one or more hardware processors, an optimizer using the plurality of predictive models to optimize the plurality of key performance parameters of the CCGT plant;
generating, via the one or more hardware processors, a second set of manipulated variables using the configured optimizer;
determining, via the one or more hardware processors, an optimal set of manipulated variables using the first set of manipulated variables and the second set of manipulated variables based on
the cause of the detected anomalies,
the determined state of the CCGT plant, and
an importance of the plurality of key performance parameters of the CCGT plant, wherein the importance is either defined by a user or obtained from the database, wherein the importance of the plurality of key performance parameters is defined based on instantaneous condition of the CCGT plant;
calculating, via the one or more hardware processors, rating points for each of the plurality of key performance parameters using the determined importance for each of the key performance parameters, for both the first set and the second set of manipulated variables;
computing, via the one or more hardware processors, a reward value utilizing the rating points calculated for first set and second set of manipulated variables;
choosing, via the one or more hardware processors, the optimal set of manipulated variables using a predefined set of conditions involving the comparison of the reward value with an upper threshold value and a lower threshold value; and
providing the optimal set of manipulated variables to optimize the operation of the CCGT plant for generation of electric power.
|