针对火电厂中混煤煤质计算不准确、配煤方案单一且考虑片面等问题，基于粒子群（PSO）优化前馈神经网络算法建立了混煤煤质预测模型；采用非支配排序多目标遗传算法（NSGAIII）建立了由最小绝对偏差型和标准差型优化指标组成的多目标优化配煤模型。对某电厂实际燃煤情况中非线性关系的煤质进行分析，并对预测煤质的不同特点和电厂机组运行特点进行分析。结果表明：基于煤质预测的多目标优化配煤方法，对煤质挥发分、灰分和灰熔点的预测精度比线性加权方法提高了4.55%，3.24%和5.60%，筛选出的6组配煤方案，兼顾了经济性、安全性和环保性，更符合配煤特点。

Aiming at the problems of inaccurate calculation of blended coal quality,single coal blending scheme and onesided consideration in thermal power plants,the particle swarm optimization (PSO) feedforward neural network algorithm was used to establish the prediction model of blended coal quality.The nondominated sorting multiobjective genetic algorithm (NSGAIII) was used to establish the multiobjective optimal coal blending model consisting of the optimization indexes of minimum absolute deviation type and standard deviation type.The coal quality of nonlinear relationships in the actual coal combustion situation of a certain power plant was analyzed,and the different characteristics of the predicted coal quality and the operating characteristics of power plant units were analyzed.The results show that compared with the linear weighting method,the prediction accuracy of volatile matter,ash and ash melting point of coal quality is improved by 4.55%,3.24% and 5.60% in the multiobjective optimal coal blending method based on coal quality prediction.The six groups of coal blending schemes are selected,which give consideration to economy,safety and environmental protection,and are more in line with the characteristics of coal blending.

TD849

上海市“科技创新行动计划”地方院校能力建设专项资助项目（19020500700）；中国华能集团有限公司2020年度科技项目（HNKJ-F2002）