基于随机森林算法的短期负荷预测研究

发布时间：2018-03-08 04:03

本文选题：短期负荷预测　切入点：电力负荷　出处：《郑州大学》2017年硕士论文　论文类型：学位论文

【摘要】：负荷预测是电力系统安全运行的基础和前提,也是电力部门安排调度计划、供电计划的重要依据。基于电力部门的市场化运行和日新月异的电力负荷需要,使得对短期负荷的预测性能提出了更高的要求。近年来在预测方向有优越表现的支持向量机(Support Vector Machine,SVM)与随机森林回归(Random Forest Regression,RFR)是两种开源算法,针对这两种算法,不同学者对它们预测优越性的比较存在很大争议,所以针对该争议本文对这两种算法的特点和预测性能进行了全面的分析,并提出自己的观点。本文首先介绍了有关短期负荷预测研究意义和目的,在对研究现状的阐述中,简单介绍了目前常用的一些短期负荷预测算法的原理和优缺点。其次在介绍负荷影响因素与特点中结合纽约市电力负荷信息进行了实例分析,得出电力负荷的周期性规律并将之应用在实际的预测中。然后对随机森林起源和算法原理进行了详细阐述,在特征变量重要性分析和相关性的分析中对实例分析做了前期准备工作。最后在实例分析中利用纽约市电力负荷和天气信息等数据,将RFR在短期负荷预测领域进行实例应用,并在预测性能的分析中与SVM算法进行全方位的综合比较,同时也提出自己的结论:RFR在短期负荷预测中预测精准度较高,泛化性能较好;在与SVM的预测性能比较中不能简单的一概而论,要结合训练样本量、特征变量还有待预测日所处时间点的不同进行具体分析。通过本文的研究,详细的了解了基于随机森林回归算法的短期负荷预测,并且在之后对于RFR与SVM预测算法的比较中起了基础指导作用。
[Abstract]:Load forecasting is the foundation and premise of the safe operation of power system, and it is also the important basis for the power department to arrange the dispatch plan and the power supply plan. In recent years, support Vector Machine (SVM) and Stochastic Forest regression Random Forest Regeneration (RFR) are two kinds of open source algorithms. There are many controversies among different scholars on the comparison of their prediction superiority, so this paper makes a comprehensive analysis of the characteristics and prediction performance of the two algorithms. In this paper, the significance and purpose of short-term load forecasting are introduced. This paper briefly introduces the principle, advantages and disadvantages of some short-term load forecasting algorithms which are commonly used at present. The periodic law of electric load is obtained and applied to actual forecasting. Then the origin of stochastic forest and the principle of algorithm are expounded in detail. In the analysis of the importance of the characteristic variables and the analysis of the correlation, the preparatory work for the case analysis is made. Finally, the RFR is applied in the field of short-term load forecasting by using the data of electricity load and weather information of New York City in the case analysis. At the same time, compared with SVM algorithm comprehensively, we also put forward the conclusion that the accuracy of forecasting is higher and the generalization performance is better in short-term load forecasting. In comparison with the prediction performance of SVM, it is not easy to generalize. In order to combine the training sample size, the characteristic variables need to be analyzed concretely by the time points of the predicted day. The short-term load forecasting based on stochastic forest regression algorithm is understood in detail, and it plays a basic role in guiding the comparison between RFR and SVM prediction algorithm.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM715;TP18

【参考文献】