Mitigating Furnace Pressure Fluctuations under Rapid Load Ramping Using a Wavelet-LSTM-PPO Based Intelligent Control Framework

Zhibin JingJianguo ShiQianpeng Hao^*Xinjian WangQiang LiMinhao Zhang

• Inner Mongolia Power (Group) Co,Ltd, Hohhot, China

Rapid load ramping in coal-fired power plants under large renewable energy integration can lead to severe furnace pressure fluctuations. Fast-changing load demands often lead to instability in the combustion process, posing risks to operational safety and control performance. To address this challenge, this study proposes a predictive and adaptive control framework based on wavelet transform, long short-term memory (LSTM) neural networks, and proximal policy optimization (PPO) reinforcement learning. First, wavelet-based multi-resolution decomposition is applied to extract key features from the fluctuating pressure signals. Then, an LSTM model is trained to forecast short-term pressure dynamics. Finally, a PPO agent learns an optimal control policy to adjust secondary air and fuel inputs in real time based on predictive feedback. The proposed method is validated using real-world data from a 600 MW supercritical boiler unit. Results demonstrate a 42.2% reduction in pressure fluctuation standard deviation, improved stability under variable load conditions, and smoother actuator response compared to traditional schemes. This study highlights the potential of combining deep learning and reinforcement learning techniques to enhance combustion stability in flexible power generation.