【目的】燃煤发热量的精准预测和评价是煤质分析和热工计算的重要基础,目前神经网络预测燃煤发热量的模型虽能有效拟合非线性关系,但存在易陷入局部最小值、收敛速度慢等问题。【方法】为精准预测煤炭在工业锅炉燃烧过程中的发热量,提出一种由仿生算法FA-GA联合优化BP神经网络的燃煤发热量预测方法。将774组燃煤锅炉常用煤的工业分析及元素分析数据进行预处理,根据平均影响值对煤质指标进行特征变量筛选,最终建立FA-GA-BP的发热量预测模型,并从误差评价指标、迭代次数等方面对优化算法寻优能力和模型预测精度进行了检验。【结果】经过特征变量筛选后模型的预测精度提升至0.9561;FA-GA联合算法与单一优化算法FA、GA、PSO相比,迭代次数显著减少,算法的全局搜索能力得到有效提升;FA-GA-BP模型与单一优化模型FA-BP、GA-BP、PSO-BP以及目前常用的发热量模型MLR和SVR相比,精度更高,相关系数可达0.9845.【结论】FA-GA算法优化BP模型,针对燃煤锅炉中来自不同地区、不同煤种的发热量预测具有良好的效果,在理论上满足了工业误差要求。改进后的燃煤发热量预测模型可为有效监测入炉煤质实时变化提供一种新方法。

【Purposes】 Accurate prediction and evaluation of coal heat generation is an important foundation for coal quality analysis and thermal engineering calculation.  The current model of neural network prediction of coal heat generation can effectively fit the nonlinear relationship, yet there are problems such as the ease to fall into the local minimum and slow convergence speed.  【Methods】 In order to accurately predict the heat generation of coal in the combustion process of industrial boilers, a coal heat generation prediction method by bionic algorithm FA-GA joint optimization BP neural net-work is proposed.  The industrial analysis and elemental analysis data of 774 groups of coal commonly used in coal-fired boilers are preprocessed, and the characteristic variables of coal quality indexes are screened according to the average impact value, and finally the heat generation prediction model of FA-GA-BP is established, and the optimization algorithm optimization ability and model prediction accuracy are examined in terms of the error evaluation indexes and the number of iterations.  【Findings】 The prediction accuracy of the model is improved to 0. 9561 after feature variable screening; the number of iterations of the joint FA-GA algorithm is significantly reduced compared with those of the single opti-mization algorithms FA, GA, and PSO, and the global search ability of the FA-GA algorithm is ef-fectively improved; the FA-GA-BP model has a higher accuracy compared with single optimization models FA-BP, GA-BP, PSO-BP, as well as the currently commonly used heat generation models MLR and SVR, and the correlation coefficient can reach 0. 9845.  【Conclusions】 The FA-GA algo-rithm optimizes the BP model with good results in predicting the heat generation from different regions and coal types in China for coal-fired boilers, which theoretically meets the industrial error requirements.  The improved coal-fired heat generation prediction model can provide a new method for effective monitor-ing of real-time changes in coal quality in the furnace.