基于MCS-MIFS与LightGBM的燃气轮机功率预测方法

预测燃气轮机的功率变化对确定机组的最优工作点有重要意义,为此,结合当前迅猛发展的人工智能技术,提出一种基于轻梯度提升机(light gradient boosting machine,LightGBM)的燃气轮机功率预测方法。将蒙特卡洛采样(MonteCarlosampling,MCS)和互信息特征选择(mutual information feature selection,MIFS)结合得到MCS-MIFS算法;利用MCS-MIFS算法对复杂的燃气轮机数据进行筛选,得到与功率最相关的属性;将上述属性进行适当处理后,作为Light GBM模型的输入来实现功率预测。以某电厂燃气轮机运行数据进行实验,结果表明通过MCS-MIFS算法选择的变量来预测燃机功率的效果优于MIFS;与梯度提升决策树(gradient boosting decision tree,GBDT)和XGBoost(eXtreme gradient boosting)两种集成模型相比,LightGBM的运行速度更快、预测精度更高、占用内存更小,在处理海量工业数据时具有优势。

Gas Turbine Power Forecasting Based on MCS-MIFS and LightGBM

Power prediction of gas turbine is significant for determining the optimal operation point of the unit.Therefore,a power forecasting method based on light gradient boosting machine(LightGBM)was proposed,based on the frontier theory research of artificial intelligence.A new algorithm named MCS-MIFS,combining Monte Carlo sampling(MCS)and mutual information feature selection(MIFS),could be used to screen the attributes most relevant to the power from massive complex data.In this way,LightGBM based on the above attributes could be achieved for power prediction.The results show that variables selected by MCS-MIFS are better for power prediction than MIFS.In addition,compared with gradient boosting decision tree(GBDT)and eXtreme gradient boosting(XGBoost),LightGBM has characteristics of faster speed,higher accuracy and less consumption,which is more conducive to processing massive industrial data.