供热工况下的汽轮机振动长周期预测与分析

以某350 MW供热机组高中压转子为研究对象，对K近邻法、随机梯度下降法、支持向量回归法和随机森林法等汽轮机高中压转子振动预测算法进行了可行性分析，并对机组首个供热期1,2号瓦轴振进行了预测。结果表明：K近邻算法、高斯径向基函数支持向量回归及随机森林算法可用于汽轮机高中压转子振动预测；在供热期振动预测中，K近邻算法易受目标参数大幅变化影响，导致预测结果偏差明显；特征参数越限极易引发支持向量回归算法精度偏差；随机森林算法具有最优的泛化能力，其在稳定工况振动预测中具有较高精度；预测结果揭示了该机组高中压转子振动水平在供热季期间不断恶化，1号瓦振幅增大20%,2号瓦振幅降低5%。

Long Period Prediction and Analysis of Steam Turbine Vibration under Heating Condition

Taking the HP IP rotor of a certain 350 MW heating unit as the research object, the feasibility analyses of K nearest neighbor, stochastic gradient descent, support vector regression and random forest algorithms for turbine HP IP rotor vibration prediction were carried out, and the vibration prediction of No.1 and No.2 bearings during the first heating period of the unit was carried out by using the above algorithms. The results show that the K nearest neighbor algorithm, Gaussian radial basis function support vector regression and random forest algorithm can be used to predict the vibration of turbine HP IP rotor. In the vibration prediction of heating period, the K nearest neighbor algorithm is susceptible to the large variation of target parameters, leading to obvious deviation of prediction results. Out of bounds characteristic parameters can easily lead to accuracy deviation of support vector regression algorithm. The stochastic forest algorithm has the optimal generalization ability and high precision in vibration prediction under stable conditions. The prediction results reveal that the vibration level of the HP IP rotor of the unit deteriorates continually during the heating season, with the amplitude of No.1 bearing increasing by 20% and that of No. 2 bearing decreasing by 5%.