基于支持向量机的锻造生产线能耗预测研究

发布时间：2018-02-01 16:46

本文关键词： 锻造生产线能耗模型多元回归支持向量机粒子群算法　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：中国的能源消费结构中,工业能源消费占到70%,伴随着能源问题的日益严峻以及能源价格的持续上涨,工业节能不仅是国家"十三五"节能减排规划的要求,更是工业企业提升综合实力的必由之路。生产线的能耗建模和预测是工业企业节能的一个重要研究方向,而机器学习算法的日趋成熟为构建性能优良的能耗模型提供了一条途径。本文首先介绍了生产线制造系统及能耗预测的相关理论,根据现代制造系统的结构和特点,总结了设备层、工艺层和系统层能耗问题的研究方法,结合能耗预测在不同领域的应用,制定了生产线能耗预测的基本框架。针对某公司铸锻中心进行能源调研工作,从工艺流程、生产表现、设备状态和物流四个方面初步选取锻造生产线的11个能耗影响因素。之后采集企业2015-2016年的能耗及相关因素数据,剔除异常样本数据、次要数据和冗余数据,筛选出生产线能耗的7个主要影响因素。在此基础上使用多元回归的方法,针对生产线的8种产品进行了能耗的多元回归建模。锻造生产线的能耗因素多样,生产过程存在着诸多不稳定因素,而机器学习算法可以利用数据模拟高维复杂系统。本文采用支持向量机方法,通过实验确定最优核函数为RBF径向基核函数,采用试凑法与五折交叉验证法完成RBF支持向量机预测模型的惩罚因子C和核函数参数g的参数选取工作。针对重点耗能设备中频感应加热炉,通过留一法结合支持向量机完成加热炉的能耗预测模型。通过引入交叉验证和自适应性变异,提高了粒子群算法的全局寻优能力,并将其应用于支持向量机超参数的优化工作,改善了生产线能耗模型的预测性能,最后通过模型精度对比,验证了改进后的能耗模型在预测精度上的优势。
[Abstract]:In the energy consumption structure of China, industrial energy consumption accounts for 70%, along with the increasingly serious energy problems and rising energy prices. Industrial energy saving is not only the requirement of the 13th Five-Year Plan, but also the only way for industrial enterprises to improve their comprehensive strength. The modeling and prediction of production line energy consumption is an important research direction for industrial enterprises. The maturation of machine learning algorithm provides a way to build energy consumption model with good performance. Firstly, this paper introduces the production line manufacturing system and the related theory of energy consumption prediction. According to the structure and characteristics of modern manufacturing system, this paper summarizes the research methods of energy consumption in equipment layer, process layer and system layer, and combines the application of energy consumption prediction in different fields. The basic framework of energy consumption prediction of production line is established. The energy research work is carried out according to the casting and forging center of a company from the process flow and the production performance. Four aspects of equipment status and logistics initially selected 11 factors affecting the energy consumption of forging production line, and then collected the energy consumption and related factors data from 2015-2016 to eliminate the abnormal sample data. The secondary data and redundant data were used to screen out the seven main influencing factors of production line energy consumption. On this basis, the method of multivariate regression was used. Multivariate regression modeling of energy consumption was carried out for 8 kinds of products in the production line. The factors of energy consumption in forging production line were various and there were many unstable factors in the production process. The machine learning algorithm can simulate the high-dimensional complex system with data. In this paper, the support vector machine method is used to determine the optimal kernel function as RBF radial basis kernel function. The penalty factor C and kernel parameter g of the prediction model of RBF support vector machine are selected by means of trial and error method and 50 fold cross verification method. The intermediate frequency induction heating furnace is used for the key energy consuming equipment. The energy consumption prediction model of the reheating furnace is completed by using a residual method and support vector machine. The global optimization ability of the particle swarm optimization algorithm is improved by introducing cross-validation and self-adaptive mutation. It is applied to the optimization of super parameters of support vector machine to improve the prediction performance of the energy consumption model of production line. Finally, the superiority of the improved energy consumption model in prediction accuracy is verified by comparing the accuracy of the model.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG318;TP18

【相似文献】