基于神经网络的铅锌烧结过程产量质量预测模型

Quantity-quality Prediction Model Based on Neural Networks in Lead-zinc Sintering Process

针对铅锌烧结过程中具有强非线性、时滞的特点,提出一种基于变学习率的烧结块产量质量神经网络预测模型。通过分析过程特性和工况参数的相关性,确定影响产量和质量的操作参数;采用普通的BP(Back Propagation,简称BP)神经网络结构,建立铅锌烧结块产量质量预测模型;在网络训练的过程中,采用变学习率的方法对BP算法进行改进,获得了满意的预测效果,该算法具有较快的收敛速度。将改进的神经网络模型进行仿真实验,结果表明,该模型具有较高的预测精度和较强的自学习功能,从而验证了方法的有效性。

There are some features of strong non-linearity and a large time delay in the lead-zinc sintering process （LZSP）, a variable-learning-rate-based back propagation neural network （BPNN） is used for predicting quantity and quality of sintering agglomeration. First, the factors influencing quantity and quality were determined by investigating the correlation of operation parameters. Then, the quantity-quali（y prediction models of agglomerations were established by applying a BPNN based on the variable-learning-rate method. In the process of training BPNN, the usual BP was improved by changing the learning rate, and satisfactory predicted results were obtained. This algorithm shows a better convergence rate. Finally, the obtained quantity-quality prediction models were applied for LZSP. Prediction results show that the proposed models possess higher accuracy and strong self-study ability. The model of quantity and quality in the LZSP is effective.