基于实数编码遗传算法的神经网络成本预测模型及其应用

Neural network cost prediction model based on real-coded genetic algorithm and its application

在生产过程中,影响产品成本的因素多而复杂,因素之间相互影响,存在耦合现象,因此准确预测成本是一个重要又难以解决的问题.通过遗传算法(GeneticAlgorithm)与误差反向传播(ErrorBackPropagation)神经网络相结合,提出了用实数编码的自适应变异遗传算法训练神经网络权重的混合算法,避免了传统神经网络易陷入局部极小的缺点.以矩阵形式表示产品成本组成,建立了产品成本组成模型,以此为基础建立了考虑成本因素之间互相影响的神经网络产品成本预测模型,并成功应用于某钢铁企业产品成本的预测,提高了预测精度.

In production process,many complex factors which influence cost affect each other and the coupling phenomenon exists,so it is important and difficult to predict the cost.By combining genetic algorithm with error back propagation neural network,a hybrid algorithm that trained neural network weight by real-coded adaptive mutation genetic algorithm is presented,and it overcomes the disadvantage that traditional neural network is easy to fall into local minima.The product cost composition is expressed by matrix,the product cost composition model is established,on the basis of the model,the product cost prediction model based on neural network is established,and the interactions among cost factors are taken into account.Furthermore,the model is successfully applied to cost prediction in some iron and steel enterprise,and the prediction precision is improved.