Scholarlink

This paper presents the development and application of a hybrid machine learning approach, using a novel application of a Deep Neural Network (DNN) model to predict gas condensates and a combined Adaptive Neuro-Fuzzy Inference System (ANFIS) and Deep Q-Learning (DQN) models to optimize turbine efficiency using gas chromatography data, marking a first study in the context of reheat gas turbine in Sarawak, Malaysia. Focused on addressing the gas condensate formation due to poor quality natural gas supply, the study utilized gas chromatography data. The DNN prediction model boasts great accuracy with its training MSE (Mean Squared Error), validation MSE and testing MSE at 0.0005669, 0.000535 and 0.02385, respectively. The DQN optimization model also shown superior accuracy with its MSE value of 0.0001467. The results indicate a 4% reduction in specific fuel consumption and a 3.9% improvement in heat rate, demonstrating the effectiveness of the proposed machine learning approach in achieving operational optimization while mitigating the adverse effects of gas condensate on gas turbine equipment.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Gas Turbine Optimization Using ANFIS and Deep Q-Learning Based on Gas Chromatography

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Bennedict Wat Lembe,
  • Lai Choo Chek,
  • Okorie Ekwe Agwu,
  • Danesh Mailvaganam,
  • Cyril Heng Hiang Jeng,
  • April Lo Chu Qiao

摘要

This paper presents the development and application of a hybrid machine learning approach, using a novel application of a Deep Neural Network (DNN) model to predict gas condensates and a combined Adaptive Neuro-Fuzzy Inference System (ANFIS) and Deep Q-Learning (DQN) models to optimize turbine efficiency using gas chromatography data, marking a first study in the context of reheat gas turbine in Sarawak, Malaysia. Focused on addressing the gas condensate formation due to poor quality natural gas supply, the study utilized gas chromatography data. The DNN prediction model boasts great accuracy with its training MSE (Mean Squared Error), validation MSE and testing MSE at 0.0005669, 0.000535 and 0.02385, respectively. The DQN optimization model also shown superior accuracy with its MSE value of 0.0001467. The results indicate a 4% reduction in specific fuel consumption and a 3.9% improvement in heat rate, demonstrating the effectiveness of the proposed machine learning approach in achieving operational optimization while mitigating the adverse effects of gas condensate on gas turbine equipment.