基于并行核径向基神经网络的软测量与控制模型

Soft sensor and control model of parallel kernel RBF NN

转炉提钒过程中存在大量多元非线性因素,难以从统计学和机理上建立各操作参数与生产目标的优化控制模型,为优化转炉的操作参数,建立了基于径向基神经网络的半钢钒含量软测量和控制模型。径向基神经网络常用于非线性回归预测和控制,但是高维的核函数矩阵运算需要花费巨大计算资源。为了缩短计算时间,本文设计了并行算法用于计算径向基网络核函数矩阵,并将它用于转炉提钒软测量和控制模型,在以MPI构建的工作站机群上执行该算法,利用实际数据验证了该算法的加速性和准确性。

In converter re-vanadium there exist a lot of diversity and non-linear factors. From the point of view of statistics and reaction mechanism, it is difficult to build up optimized control model between operating parameters and product goals. A soft sensor and control model for forecast vanadium output by RBF NN is presented to optimize operating parameters of converter. Radial basis function neural network （RBF NN） is frequently used for non-linear regression prediction and control, but the kernel matrix computation for high dimensional data source requires heavy computing power. To shorten the computing time, a parallel algorithm to compute the kernel function matrix of RBF NN was designed and applied to the soft sensor and control model of converter re-vanadium. The algorithm was implemented on a cluster of computing workstations using MPI. Finally, the experimental data prove that the algorithm can speed up the computation and its result is accurate.