Steam Power Plant Optimization

Abstract

The study presents a framework for utilizing supervised Artificial Intelligence - Machine Learning (AI-ML) to optimize and enhance the operations of steam power plants. We focus on the steam power plant’s boiler subsystem, which is a key component, as a case study. Other significant subsystems of power plants include the condenser, turbine, feed-water, and other auxiliary subsystems. Then, we calibrated our simulation model based on a data-set from a real power plant, in order to obtain a realistic synthetic dataset, and used it for training different types of AI-ML models. Our objective is to create a data-set that can be included in an AI-ML algorithm using this simulation model. In our training set, the boiler’s controllable inputs were selected as features, while steam temperature and pressure were designated as multi-dimensional output values. Because we can readily change the input parameters without involving or affecting the actual power plant, the simulation data-set was selected. Comparing the outputs of the actual power plant with the data-set’s reality as a result. The suggested framework and an explanation of the machine learning approach are covered in this study. Using various regression analyses, the data set will be subjected to further processing for training and testing. The data generated part is focused on three machine learning algorithms: random forests, decision trees, and linear regression. Moreover, deep learning will also be used on this data-set, where we will focus on artificial neural networks. Validation was performed on two held-out data-sets, checking the capability of the different models of predicting the outputs of the calibrated simulator, as well as the ones of the actual power plant. Analyzing the effectiveness of various algorithms using the R2-score, mean squared error, root mean squared error and mean absoulute error.

Date of Award	Apr 2023
Original language	American English
Supervisor	Ernesto Damiani (Supervisor)