Eventos Anais de eventos
ENCIT 2020
18th Brazilian Congress of Thermal Sciences and Engineering
Big Data Clustering Model for the Identification of a Thermal Power Plant Operating Patterns
Submission Author:
Jéssica Duarte , Procurando endereço...
Co-Authors:
Jéssica Duarte, Lara Werncke Vieira, Augusto Delavald Marques, Paulo Smith Schneider
Presenter: Jéssica Duarte
doi://10.26678/ABCM.ENCIT2020.CIT20-0356
Abstract
Thermal power industry is characterized by complex and challenging processes, dependent of numerous variables. Its information is accessed by a Distributed Control System (DCS) which generates thousands of data that are difficult to analyze together. This paper proposes to recognize the different patterns that occur on a thermal power plant operation, by means of unsupervised machine learning methods based on historical data. The proposed methodology in this paper is applied to an industrial data set from a 360 MW coal-fired thermal power plant located at Ceará, in Brazil. Initially, the methodology is applied to 40 selected parameters from the steam generator and mills operation. The studied dataset has its dimensionality and redundancy reduced by principal component analysis (PCA), defining a lower-dimensional space proper for clustering while preserving most of its variance. Hence, the K-means clustering method identifies operating points groups according to their degree of similarity. The appropriate clusters number is defined by an analysis with the average silhouette coefficient, which measures the clusters consistency. Following the definition of the clusters, its parameters values and distribution are evaluated in order to verify the consistency of the results. For the case studied, two analysis were evaluated, considering the initial 40 parameters and considering a selection of 29 parameters. The latter’s results presented more conformity to the power plant’s operation, being described by a 2 clusters analysis overall or by a 10 clusters analysis, for refined observations. The results indicate that the method was able to distinguish different operation arrangements.
Keywords
Power plant operation, Operation patterns, Operation parameters, k-means clustering, Principal component analysis (PCA)
DOWNLOAD PDF VIEW PRESENTATION
