摘要
在研究加工参数对颗粒饲料产品质量所产生的重要影响基础上,建立以生产近似模型和多目标优化算法为核心的加工参数优化设计方法,由此面向产品质量、生产能力等指标提供针对性全局优化方案。基于加工参数-质量指标体系,以人工神经网络为基础构建参数-指标近似模型;使用多学科优化软件i SIGHT,基于改进非支配排序遗传算法(non-dominated sorting genetic algorithmII,NSGA-II)对各目标进行优化。应用实例分析表明,由实际企业自身要求构建的优化参数-质量指标体系为基础建立近似模型,获得的近似值与实际值间呈显著正相关,平均决定系数值R2达0.94以上,结合验证误差指标可知近似模型满足优化使用要求;通过设计可能性历史图、二维Pareto前沿图等分析多目标之间存在的制约关系,根据优化数值结果给出了全局优化方案:环模长径比为17,模孔直径为3.0 mm,锤片粉碎机筛片孔径分别为2.0、1.5 mm,调质温度76.2℃;对优化结果进行验证,可知在该企业生产条件下生产率及PDI指标优化幅度分别达到了22.04%和6.04%,展示出一定程度的优化效果;进一步在其他生产企业应用本文方法,验证结果显示该企业采用的PDI、生产率、含粉率等指标的优化幅度分别达到了4.03%、33.93%和28.49%,进而可证明该方法的有效性及实用性。文章优化设计方法为颗粒饲料产品质量控制提供了一种新思路,可作为饲料行业高效低耗生产的理论参考及实用性工具。
By a method of mechanical compression and forced through a die, pellet feed are obtained with advantages such as avoidance of picky eaters of anita., better homogeneity and other characteris- tics, which had become the main feed products application form. Studies show that in the pellet feed production, processing parameters have a significant impact on pellet teed products quality and animal performance, so the optimization of pellet feed processing parameters is very necessary and of great practical significance to feed industry development. In this article, a multiple objectives optimization design was constructed with the approximate model of artificial neural network as well as the non-dom- inated sorting genetic algorithm as the optimization method, with help of iSIGHT software, which could provide efficient solution package of pelleting processing parameters for industry. The followed application in a practice feed mill of this multi-ple objectives optimization design method showed the efficiency and practicality： with MAPE and other error indicators to further evalu- ate the approximate performance, the data show that after the optimization of PSO, the BPNN approximation model has achieved a high approximation accuracy which can meet the requirements of the pellet feed processing process. Basing on applying situation, several key parameters in pellet processing were decided to be the optimized parameters meeting with actual requirements of feed mill, including mash size of screen （big/small）, conditioning temperature, length-diameter ratio of ring die and diameter of die orifice. Indexes to measure pellet quality and production were pellet durability index, hardness and productivity. Eventually, basing on actual production data, it can be found that the optimization of the objectives is considerable, evi- denced by the optimized range of productivity and PDI reached 22.04% and 6.04%, which can provide users with meaningful reference. On the other hand, verification test are conducted in another feed mill to prove the feasibility of the design method. After applied the method using actual production da- ta from this feed mill, the comparison consequences after the optimization shows that there are respec- tively optimized range of 4.03%, 33.93% and 28.49% towards objectives including PDI, productivity and powder content ratio, targeting to this very feed mill. The preferable optimized results from another feed mill verified the efficiency and feasibility of this design method. Lastly, users should select unique target processing parameters and objectives basing on their own actual need, and then collect- ing actual production data to train the approximate model until fulfilling sufficient accuracy that can describe the genuine process relationship of the target mill; thus finish the multi-objectives optimiza- tion design. Future research should be continued for balance model and optimization algorithm astrin- gency which can be applied in a certain extent to global optimization for conflict targets thus achieve the design goal and better benefit the feed industry practitioners.
作者
陈啸
孔丹丹
方鹏
王红英
Chen Xiao Kong Dandan Fang Peng Wang Hongying
出处
《饲料工业》
北大核心
2017年第19期7-13,共7页
Feed Industry
基金
公益性行业(农业)科研专项[201203015]
关键词
颗粒饲料
产品质量
加工参数
近似模型
多目标优化算法
pellet feed
production quality
processing parameters
approximate model
multiple objectives optimization algorithm
