Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
VERIFYING EXPLAINABILITY OF STEAM GENERATOR EFFICIENCY PREDICTION WITH SHAP VALUE INTERPRETATION
Submission Author:
Lara Werncke Vieira , RS
Co-Authors:
Lara Werncke Vieira, Augusto Delavald Marques, Jéssica Duarte, Rodrigo Ghiorzzi Donni, Paulo Smith Schneider
Presenter: Lara Werncke Vieira
doi://10.26678/ABCM.COBEM2021.COB2021-1152
Abstract
Complex engineering systems, such as power plants, deliver their best performance when operating along with a designed range of some priority parameters. Predicting a power plant efficiency requires more flexible models and thus more adaptable to the complex behavior of the real world, such as non-linear relationships and interactions between variables. Machine Learning (ML) is usually seen as an option for powerful algorithms with high accuracy but without intelligibility. This paper aims to predict the steam generator efficiency of a coal-fired power plant and quantify the average contribution that each feature brings to the prediction made by the model. In this regard, Deep Learning (DL) techniques are applied to steam generation efficiency prediction, while Shapley Additive Explantions (SHAP) game theoretic approach is used to explain the outputs of the data-driven model. As a result, it quantifies each model input's impact and analyzes the model decisions, helping to build better models and guide the operators' decisions. The impact of varying parameters that influence steam generator efficiency plays a vital role in the daily operational management of power systems without the need to choose between accuracy and explainability on the models.
Keywords
SHAP value interpretation, Artificial neural networks (ANN), coal-fired power plant, Explainability
