基于改进BP神经网络的电网物资需求预测研究

发布时间：2018-03-20 02:15

本文选题：BP神经网络　切入点：电网物资　出处：《华北电力大学》2014年硕士论文　论文类型：学位论文

【摘要】：随着国家电网公司“三集五大’’建设的全力推进,物资集约化管理越来越受到重视,电网物资的一级集中采购范围也越来越广,这也就意味着国家电网公司对各网省电力公司所需申报的年度物资需求计划的要求也越来越严格。而现有的年度物资需求计划提报信息主要还是依据项目资料以及人为经验,缺乏一定的科学依据,准确度也难以满足要求,因此开展电网物资的需求预测有着重要的实际意义。另外,对于电网物资的采购管理来说,需求预测的开展对合理安排物资招标、提高资金利用率以及减少物资积压等也有着重要的意义。本文深入分析了电网物资的需求特性,按照物资的用途维度详细阐述了项目类物资以及非项目类物资的需求特性,确立了以基建项目为单位,开展电网物资需求预测的研究思路。同时,通过对现有的物资需求预测方法的分析比选,确定了应用BP神经网络模型来进行电网物资需求预测的思路,并详细阐述了BP神经网络的模型结构、学习参数选择以及学习算法。并在对标准BP神经网络算法分析的基础上,利用SCG算法以及遗传算法分别对其训练算法以及初始权值、阈值的选取进行了改进,构建了基于改进BP神经网络的电网物资需求预测模型,并给出了详细的网络设计。最后以110kV新建线路工程物资-钢芯铝绞线需求预测为例,进行了实例计算,验证了模型的科学性和有效性,并对模型的适用性进行了深入探讨。
[Abstract]:Along with the national Power Grid Corp "three sets of five 'efforts to promote the construction, material intensive management has been paid more and more attention, the level of power supplies centralized procurement range more widely, which means that the national Power Grid Corp of the annual material demand plan each provincial power network company to declare more and more strict. But the existing annual MRP report information is mainly based on the project information and human experience, lack of scientific basis, the accuracy can not meet the requirements, so to carry out the grid material demand forecast has practical need. In addition, for the procurement management of power supplies, the demand forecast of reasonable development arrange material bidding, improve capital utilization and reduce the backlog of material also has important significance.
This paper analyzes the demand characteristics of the power grid materials, expounds the demand characteristics of items and non project materials according to the dimension of materials, and establishes the research train of thought for the prediction of power grid material demand based on infrastructure projects.
At the same time, through the analysis of the existing material demand forecasting method selection, the application of BP neural network model to forecast the demand of power grid construction ideas, and expounds the model structure of BP neural network, learning parameter selection and learning algorithms. And based on the analysis of the standard BP neural network algorithm, respectively the training algorithm and the initial weights by SCG algorithm and genetic algorithm, the threshold selection is improved, build a prediction model of grid material demand based on improved BP neural network, and gives the detailed design of the network. Finally, with the newly built 110kV line engineering materials - ACSR demand forecast for example, was calculated. Verify the scientificity and validity of the model, and the applicability of the model are discussed.

【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP183;F274;F426.61

【参考文献】