基于神经网络的水泥生料配料多目标优化设计方法

Multi-objective optimization design method of control system in cement raw materials blending process based on neural network

为了降低生料成分的不确定性给水泥生料质量控制系统带来的影响,提出了率值补偿的控制策略.分别为三率值创建目标函数,并利用状态空间搜索策略解决多目标优化问题.针对初始样本空间不能覆盖所有样本的问题,提出了基于神经网络的估算模型,对初始样本空间进行拓扑.通过估价函数对状态空间中的状态量进行评价,得到最优的率值状态量;根据率值对原料配比进行调整,最后使率值偏差得到补偿,同时使给配比造成的波动最小.工业实验结果表明,生料的质量合格率由原来的30%提高到50%,该系统能有效地对配料过程进行优化控制.证明了基于神经网络的状态空间搜索策略为水泥生料配料多目标寻优问题提供了一种可行的方法.

In order to reduce the influence of the uncertainty of raw material components on the control system of cement raw material quality,a control strategy is proposed based on modulus value compensation.Goal functions are created for three modulus values separately,and multi-objective optimization problem is solved with the strategy of state-space search.Aiming at the problem that the initial sample space cannot cover all of the samples,the estimation model based on neural network is brought forward for initial sample space topology.Through evaluating the quantity of state with evaluation function,the optimal quantity of state of modulus value is obtained.Then the raw material configuration is adjusted according to the modulus value,and finally the deviation of modulus value is compensated;the smallest fluctuation to the mixture ratio is made at the same time.Industrial experimental results show that the pass rate of quality is increased from 30% to 50%,which proves that the system can control raw material blending process effectively and efficiently,and thus proves that the state-space search strategy based on neural network is available for the problem of multi-objective optimization of cement raw material blending.