【摘要】 为解决燃气-蒸汽联合循环机组热效率的退化精准预测问题，该文提出一种基于随机森林与向量回归及自回归神经网络的综合机器学习方法。该方法采用随机森林(random forest,RF)对采集数据进行标准工况的折算处理，通过支持向量回归(support vector regression,SVR)模拟汽轮机在健康状态下的运行参数并建立退化指标，并构建自回归神经网络(auto-regression neural net,AR-Net)汽轮机热效率退化预测模型并用来预测热效率退化趋势。结果表明，RF模型有效减轻了环境波动的影响；SVR模型准确预测了汽轮机在健康状态下的运行参数；AR-Net模型则准确预测了长时间跨度上的退化趋势。可知，该模型通过整合历史与实时数据，能够有效捕捉汽轮机热效率退化趋势，可为汽轮机的预测性维护提供一定指导。更多还原

【Abstract】 The long-term operation of steam turbines inevitably leads to thermal efficiency degradation. Accurately predicting the trend of steam turbine thermal efficiency degradation is crucial for optimizing maintenance cycles and enhancing operational economy and safety. Given the complexity of steam turbine operation, especially the influence of variable working conditions and environmental factors, this study proposes a comprehensive machine learning method for predicting the degradation of steam turbine thermal efficiency. This method integrates random forest(RF), support vector regression(SVR), and autoregressive neural networks(AR-Net). Specifically, RF is used to standardize the collected data, SVR is employed to simulate the operational parameters under healthy conditions and establish a degradation index, and finally, AR-Net is applied to construct a predictive model for the trend of efficiency degradation. The results indicate that the RF model effectively mitigates the impact of environmental fluctuations, the SVR model accurately predicts the operational parameters under healthy conditions, and the AR-Net model precisely forecasts the long-term degradation trend. In summary, by integrating historical and real-time data, the model effectively captures the trend of thermal efficiency degradation, providing guidance for predictive maintenance of steam turbines.更多还原