基于PSO-Elman神经网络的燃气轮机压力脉动预测

谷泽文¹，丁阳²，苗玉彬^1*，李明²

1.上海交通大学机械与动力工程学院，上海  200240；

2.华电电力科学研究院有限公司，浙江 杭州  310013

摘要：为准确预测燃气轮机燃烧调整过程中的燃烧不稳定现象，提出一种融合信号预处理、敏感性分析与智能算法的混合建模框架；采用经验模态分解（empirical mode decomposition，EMD）与小波阈值去噪相结合的方法对敏感频段压力脉动信号进行预处理，以抑制噪声干扰；通过多维度敏感性分析，筛选影响燃烧状态的关键输入特征；构建基于粒子群优化（particle swarm optimization，PSO）算法的Elman（PSO-Elman）递归神经网络预测模型，仿真分析燃气轮机在不同负荷下的压力脉动。结果表明：与传统Elman神经网络相比，PSO-Elman模型在功率为270 MW时的预测均方误差降低了63.1%，平均绝对百分比误差降低了28.4%；与未去噪处理的PSO-Elman模型相比，对压力脉冲信号进行去噪预处理后的PSO-Elman神经网络预测模型预测的均方误差降低了70.8%，表明该模型能够有效追踪燃烧过程中压力脉动的动态演变趋势，可作为实现燃烧状态实时监测与优化调整的理论工具和技术途径。

关键词：燃烧调整；EMD；PSO算法；Elman神经网络；燃烧稳定性

Gas turbine pressure fluctuation prediction based on PSO-Elman neural network

GU Zewen¹, DING Yang², MIAO Yubin^1*, LI Ming²

1.School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China;

2.Huadian Electric Power Research Institute Co., Ltd., Hangzhou 310013, China

Abstract: To accurately predict combustion instability phenomena during the combustion adjustment process of gas turbines, a hybrid modeling framework integrating signal preprocessing, sensitivity analysis, and intelligent algorithms is proposed. The method combining empirical mode decomposition (EMD) and wavelet threshold denoising is adopted to preprocess the pressure pulsation signals in sensitive frequency bands to suppress noise interference. Through multi-dimensional sensitivity analysis, key input features affecting combustion status are identified. An Elman recurrent neural network prediction model based on particle swarm optimization (PSO) is constructed to simulate and analyze pressure pulsations in gas turbines under different power loads. The results show that compared with the traditional Elman network, the PSO-Elman model reduces the mean squared error by 63.1% and the mean absolute percentage error by 28.4% at a power load of 270 MW. Compared with the PSO-Elman model without denoising, its mean squared error is reduced by 70.8%. This model can effectively track the dynamic evolution trend of pressure pulsations during combustion and can serve as a theoretical tool and technical approach for real-time monitoring and optimization adjustment of combustion status.

Keywords: combustion adjustment; EMD; PSO algorithm; Elman neural network; combustion stability