基于量子微粒群的BPNN在转炉炼钢静态模型中的应用

摘要

针对转炉炼钢静态模型终点命中率较低的问题,首先分析了影响转炉炼钢终点命中率的各种因素,确定了BP神经网络(BPNN)的拓扑结构,并依此建立了转炉炼钢静态模型.然后把量子微粒群算法(QPSO)应用于BP网络的学习中,并比较了QPSO、基本微粒群优化算法(PSO)、梯度下降法的学习性能.最后,基于某炼钢厂的历史数据进行了仿真实验,比较了三种BP网络学习算法下的炼钢终点命中率.研究结果表明,该研究提高了转炉炼钢静态模型的终点C含量和温度预测精度.%For the problem of low hit rate of the BOF endpoint based on static model, the factors that affect the hit rate of the BOF endpoint was firstly analyzed, topologies of the BP neural network (BPNN) were determined, the static BOF model was established.Then the quantum particle swarm optimization (QPSO) was used in the study of BP network, and the learning performance of QPSO, the basic particle swarm optimization (PSO), gradient descent was compared.Finally, experiment based on historical data of a steel plant was simulated, the hit rate of the BOF endpoint was compared under three types of BP network learning algorithm.The results indicate that the analysis improves prediction accuracy of the converter end C content and temperature.