基于PCA-BP神经网络的转炉终点磷含量预报模型

Prediction model of endpoint phosphorus content of converter based on PCA-BP neural network

转炉炼钢终点控制是转炉吹炼后期的重要操作,为了更加准确地预报转炉炼钢终点磷含量,选取影响终点磷含量的13个工艺参数,然后采用灰色关联度分析和主成分分析(PCA)处理得到输入参数,通过比较不同隐含层节点个数的预报结果的均方误差值确定隐含层节点个数,结合可变学习速率的BP算法,基于PCA-BP神经网络建立了转炉终点磷含量预报模型,并对Q235钢种实际生产数据代入模型进行仿真。通过与传统BP、PCABP神经网络以及小波神经网络建立的模型结果进行对比,表明算法优化后的PCA-BP神经网络的终点命中率更高,该模型实现预测转炉终点磷质量分数在误差范围±0.004%、±0.008%和±0.01%内,命中率分别达到44%、86%和96%。

The endpoint control of converter steelmaking is an important operation in the later stage of converter blowing.In order to predict the end point temperature of converter steelmaking more accurately,13 process parameters that affect the endpoint phosphorus content were selected,and then the input parameters were obtained by grey correlation analysis and principal component analysis(PCA).The number of hidden layer nodes was determined by comparing the mean square error of the prediction results of different number of hidden layer nodes.Combined the BP algorithm with variable learning rate,the prediction model of converter endpoint phosphorus content was established based on PCA-BP neural network,and the actual production data of Q235 steel was substituted into the model for simulation.Compared with the results of the model established by the traditional BP,PCA-BP neural networks and wavelet neural network,it is indicated that the endpoint hit rate of the optimized PCA-BP algorithm neural network is higher,and the hit rate of endpoint phosphorus content is 44%,86%and 96%respectively when prediction errors are within±0.004%,±0.008%and±0.01%.