Abstract: BP ( Back Propagation) algorithm and genetic algorithm are combined into hybrid genetic algorithm of which the algorithm steps are first to locate a favorable searching region by genetic algorithm, then to search optimal coefficients in the located region by BP algorithm. An 100 t arc furnace end aim temperature neural network predictive model is trained respectively by genetic algorithm and hybrid genetic algorithm in this paper. The simulation results show that the hybrid genetic algonthm has faster convergence speed and higher predictive precision, as aim temperature precision is ± 2 ℃ , ±4 ℃ , ±6 ℃ and ± 8 ℃ , the percentage of hits for aim temperature by standard genetic algorithm is respectively 75% , 82% , 86% and 92% while that by hybrid genetic algorithm is respectively 80% ,88% ,90% and 96%.

Key words: Hybrid Genetic Algorithm, Neural Network, Predictive Model, Arc Furnace, End Aim Temperature

摘要： BP (Baek Propagation)算法和遗传算法相结合的混合训练方法步骤为:首先用遗传算法定位出一个较好的搜索空间,然后采用BP算法在这个小的解空间中搜索出最优解。分别用遗传算法和混合遗传算法训练100 t电弧炉终点温度神经网络预报模型。仿真结果表明:混合遗传算法有更快的收敛速度和更高的预报命中率。当目标温度的精度范围为±2℃、±4℃、±6℃和±8℃时,BP算法的温度命中率分别为75%、82%、86%和92%,混合遗传算法的温度命中率分别为80%、88%、90%和96%。

关键词: 混合遗传算法, 神经网络, 预报模型, 电弧炉, 终点目标温度