炼焦生产过程综合生产指标的改进神经网络预测方法

An improved neural network method for the prediction of comprehensive production indices in coking process

针对炼焦生产过程综合生产指标(焦炭质量、产量和焦炉能耗)检测的严重滞后问题,提出一种改进BP神经网络预测方法.首先基于相关过程参数的主元分析和灰色关联分析,确定出预测模型的输入输出变量;然后采用基于改进差分进化算法的BP神经网络建立预测模型,并与基本BP神经网络预测模型进行比较;最后,对改进BP神经网络预测模型进行了验证.实验结果表明,改进BP神经网络预测模型具有较快的收敛速度和较高的预测精度,模型的预测效果可以满足生产工艺要求.

A prediction method based on the improved back propagation（BP） neural network is proposed to solve the problem of large time-delay in the detection of the comprehensive production indices （quality and quantity of coke, and energy consumption of coke oven） in the coking process. First, the input and output variables of the prediction models are determined by analyzing the process mechanism correlation between process parameters based on principal components analysis and grey relational analysis. Then, the BP neural network based on an improved differential evolution algorithm is applied to establish prediction models, which are compared with the basic BP neural network prediction models. Finally, the prediction models are verified. Simulation results show that the proposed prediction models provide a better convergence rate and higher prediction accuracy, and the prediction effect of the obtained models satisfy the technological requirements.