Session: 01-04: AI for Energy Sustainability IV

Paper Number: 130585

130585 - Efficiency Optimization in Thermal Generation Plants: Implementation of a Functional Machine Learning Strategy Based on SCADA Data Processing 

Abstract: 

This article presents a case of an effective implementation of a machine learning model based on SCADA data in a thermal power generation plant, with the purpose of improving and maintaining energy efficiency while also providing a support tool for maintenance management.

In the context of thermal power generation plants, optimizing efficiency and maintenance management is essential to ensure reliable and profitable energy production. This article provides a comprehensive study of the successful implementation of a machine learning (ML) model based on SCADA data from a thermal power generation plant with a 16 MW steam turbine and a 22 MW synchronous generator. The implementation process followed the seven-step ML framework and was based on a dataset that included approximately 80 variables associated with the turbine generator set operation at a cadence of 1-second, during a periodo of one month.

Data collection and preprocessing played a critical role in preparing the data for analysis. Cleaning, normalization, and the selection of relevant variables were carried out to ensure the quality of input data for the ML model. Additionally, a specialized process was implemented to create the routines of recurrent data processing (export shedulling – data processing – import schedulling and calculated data tag writting) in the SCADA system, further enriching the available data.

The core of this study lies in formulating the ML model with the most suitable method for the physical problem, using variables that represent the system's state. What distinguishes the approach taken is the rather unique ability to perform both the calculation of composite variables and the ML algorithm online. This enables real-time decision-making based on pattern identification and correlations in terms of representative variables with physical significance from an engineering perspective.

The model's validation demonstrated its effectiveness in establishing performance metrics for making significant improvements in system efficiency and maintenance management. This article underscores the importance of combining engineering expertise with machine learning capabilities to address challenges in complex systems, rather than limiting the latter to structuring solutions based on pre-established correlations.

Presenting Author: David Nino PROACTIVE AND INNOVATIVE SERVICES

Presenting Author Biography: Electrical and Mechanical Engineer with MsC in Electrical Engineering and a PG diploma in Machine Learning (ML) and Artificial Intelligence (AI). More than 30 years of experience in different areas related with electrical maintenance, including: power systems analysis, electrical system failure analysis, electrical testing, predictive maintenance, electrical equipment installation and start up, field maintenance of power transformers, field maintence of power generators and MV motors, electrical protection systems and automation and control. Special interest and currently involved in activities regarding energy efficiency improvement, power quality and energy storage solutions (BESS systems, active filters and reactive compensation systems), industrial automation involving IOT, plant reliability and security, and the application of fuzzy logic and machine learning techniques in maintenance to identify and analyze industrial machinery & systems health representative variables, as well as to establish acceptance criterion for test & measurement results.

Authors:

David Nino PROACTIVE AND INNOVATIVE SERVICES
Juan Andrés Reverend Lizcano PROACTIVE AND INNOVATIVE SERVICES SAS - P&I SERVICES SAS