基于PSO和ANN的采选品位智能约束优化  

Intelligent Constrained Optimization of Mining and Ore-Dressing Grades based on PSO and ANN

作　　者：贺勇[1] 廖诺[1] 莫赞[1] 

机构地区：[1]广东工业大学管理学院,广州510520

出　　处：《系统管理学报》2014年第5期737-742,754,共7页Journal of Systems & Management

基　　金：国家自然科学基金资助项目(71303061;71301030;71171062);教育部人文社科研究项目(11YJCZH057)

摘　　要：根据铁矿采选生产过程,建立了以经济效益为目标函数,资源利用率和精矿产量为约束条件,截止品位和入选品位为决策变量的非线性约束优化模型,将粒子群算法和神经网络集成构成PSO-ANN算法来搜索最优品位组合。PSO-ANN算法包括内外两层：外层采用PSO作为搜索算法,采用基于可行性规则的约束处理技术,更新粒子群个体最优位置和全局最优位置,引导粒子朝最优解方向进行搜索;内层是REG模型、BP神经网络及RBF网络,实现粒子（截止品位和入选品位）到损失率、选矿金属回收率和采选成本之间的映射关系,进而计算资源利用率、精矿总量和净收益。以大冶铁矿为例,研究表明：2008-01~06,最优截止品位为17.5%,入选品位为45.4%,与现行方案相比,其资源利用率提高2%,精矿量增加1.34万t,总现值增加1 125万元。该方法为金属铁矿的品位优化提供了一个全新的思路,具有广泛的应用前景。We build a nonlinear constrained optimization model according to the metal mining process, where the objective function is the economic benefit, the two constraints are resource utilization rate and output of concentrate, and decision variables are cut-off grade and dressing-grade. Particle swarm optimization （PSO） algorithm and artificial neural network （ANN） are integrated as PSO-ANN algorithm to optimize the cut-off grade and dressing-grade. The outer layer of PSO-ANN uses PSO algorithm to carry out global search, the constraint handling techniques of feasibility-based rules are used to update the pbest and the gbest, and to guide the particles toward to the optimum. The inner layer uses REG model, BPNN and RBFNN to calculate the loss rate, ore-dressing metal recovery rate, and cost, respectively, in order to calculate the resource utilization rate, output of concentrate and economic benefit of each particle. Taking Daye Iron Mine as an example, we show that the optimal cut-off grade and dressing-grade during January 2008 to June 2008 are 17.5%, and 45. 4%, respectively. Compared with the current value of grades, the optimal value improves the resource utilization rate by 2%, increases the output of concentrate by 13400 tons, and improves the net present value by 11.25 million Yuan. The proposed intelligent method provides a brand-new idea to optimize grades, and may be broadly applied in metal mines.

关 键 词：采选品位 智能约束优化 粒子群算法 人工神经网络 

分 类 号：N945.15[自然科学总论—系统科学]