Performance Prediction and Optimization of the Air‐Cooled Condenser in a Large‐Scale Power Plant Using Machine Learning

• Published in: Energy technology (Weinheim, Germany) Germany), 2021-07, Vol.9 (7), p.n/a
• Main Authors: Chen, Heng, Peng, Weike, Nie, Chunming, Xu, Gang, Lei, Jing
• Format: Article
• Language: English
• Subjects: air-cooled condensers; Algorithms; Artificial intelligence; Artificial neural networks; Coal-fired power plants; Learning theory; Machine learning; Mathematical models; operation optimization; Optimization; Parameter identification; Performance prediction; Power consumption; Power plants
• ISSN: 2194-4288; 2194-4296
• DOI: 10.1002/ente.202100045

An artificial intelligence approach using machine learning (ML) is applied to predict and optimize the operational performance of the air‐cooled condenser (ACC) in a large‐scale power plant. The in situ data of one whole year are collected from a typical coal‐fired power plant equipped with an ACC. The ML models for predicting the gross power output and the ACC power consumption are established using the artificial neural network (ANN) and random forest (RF) algorithms, and appropriate parameters are determined for the ML models. As the RF models with lower errors are demonstrated to be superior to the ANN models, the RF models are selected. Then the RF models are simplified by identifying the primary input parameters and exploiting fewer input parameters. Based on the derived RF models, the operational performance of the ACC is examined when the total rotational frequency of air fans varies, and the maximum net power output is obtained with the optimal total rotational frequency of fans. Due to the optimization, the average net power output of the plant can be promoted by 0.9774 MW, and the mean of the total rotational frequency of fans increases by 11.40 Hz. Therefore, the proposed approach is beneficial and advantageous. Supervised machine learning is adopted to improve the air‐cooled condenser performance in a large‐scale power plant based on historical data. Machine learning models are established to predict the operating characteristic of the air‐cooled condenser, and performance optimization is achieved by adjusting the total rotational frequency of fans according to the prediction results.