考虑曲线特征和多影响因素的售电量预测关键技术研究与应用
发布时间：2018-12-25 15:34
【摘要】：售电量是电网企业重要的经济考核指标,月度售电量预测工作是电网企业营销部门一项重要的日常工作,准确的进行月度售电量预测可以为电网企业提供营销决策支持,对制定增供扩销计划、开展电能替代、实施有序用电方案、提升客户服务品质等具有重要意义。目前各电网企业月度售电量预测多采用对比分析、结构分析、回归分析、神经网络等方法。这些方法可在一定程度上对售电量进行预测,但对国家电网公司整体售电量预测精度并不是很理想,其主要原因是没有考虑国家电网公司各地区售电量曲线的不同特征,只利用一种预测算法对多地区的售电量进行预测,这样必然会导致预测精度不高。为解决上述问题,本文提出了两种方法。一种是基于历史曲线的售电量预测方法。根据国家电网公司及下属27家省(市)公司售电量曲线在时域和频域下的特征,对27家省(市)公司进行聚类。对不同类别的省(市)公司,根据售电量曲线特征与预测算法(SVM回归、BP神经网络、ARIMA等)的适配性,选择相应的预测方法,对同一类别内的省(市)公司采用同一种预测算法。在基于历史曲线的售电量预测的基础上,本文将天气、经济、节假日和社会事件等影响因素纳入考虑,建立基于SVM回归的售电量预测修正模型,根据影响因素的月度售电量预测修正模型,进一步提高预测精度。另一种方法是考虑春节因素的售电量调整方法,该方法首先利用历史年第一季度每月售电量占季度比重和第一季度每月1日距离当年春节的天数建立函数关系,天数为输入,占季度比为输出,利用得到的一元函数预测1、2、3月份售电量占季度比,进而根据预测的占季度比及调整前售电量预测值得到基于春节因素调整后1、2、3月份预测值。利用本文的预测方法,以国家电网公司2010年至2014年的售电量数据作为历史数据,对国家电网公司2015年每月的售电量进行预测,然后和实际的2015年售电量比较,预测平均误差为1.78%,结果表明,本文提出的售电量预测方法可靠,有效,且精度较高。
[Abstract]:Electricity sales is an important economic assessment index for power grid enterprises. Monthly electricity sales forecasting is an important daily work of power grid enterprise marketing department. Accurate monthly electricity sales prediction can provide marketing decision support for power grid enterprises. It is of great significance to make the plan of increasing supply and expanding sales, to carry out electric energy substitution, to carry out orderly power consumption scheme, and to improve the quality of customer service. At present, the monthly electricity sales forecast of power grid enterprises mostly adopts the methods of comparative analysis, structure analysis, regression analysis, neural network and so on. These methods can be used to predict the electricity sales to a certain extent, but the accuracy of the overall electricity sales prediction of the State Grid Company is not very good. The main reason is that the different characteristics of the electricity sales curves in the various regions of the State Grid Company are not considered. Only one prediction algorithm is used to predict the electricity sales in many areas, which will inevitably lead to the low accuracy of the prediction. In order to solve the above problems, two methods are proposed in this paper. One is the forecasting method of electricity sales based on historical curve. According to the characteristics of the electricity sales curve of the State Grid Company and 27 provincial (municipal) companies in the time domain and the frequency domain, 27 provincial (municipal) companies were clustered. According to the characteristics of the sales curve and the adaptability of the prediction algorithm (SVM regression, BP neural network, ARIMA etc.), the corresponding forecasting methods are selected for different kinds of provincial (municipal) companies. The same prediction algorithm is used for provincial (municipal) companies in the same category. On the basis of forecasting electricity sales based on historical curve, this paper takes weather, economy, holidays and social events into account, and establishes a revised model of electricity sales forecasting based on SVM regression. The forecast accuracy is further improved according to the monthly electricity sales forecast correction model based on the influencing factors. Another method is to take into account the factors of the Spring Festival to adjust electricity sales. Firstly, the method uses the proportion of electricity sales per month in the first quarter of a historical year and the number of days between the first quarter and the Spring Festival in the first quarter to establish a functional relationship. The number of days is input. The ratio of quarter to quarter is output, and the one-variable function is used to forecast the quarterly ratio of electricity sales in March, and then according to the predicted quarterly ratio and the forecast value of electricity sales before adjustment, the forecast value for March is based on the adjustment of Spring Festival factor. Using the forecasting method of this paper, taking the electricity sales data of State Grid Company from 2010 to 2014 as historical data, this paper forecasts the monthly electricity sales of State Grid Company in 2015, and then compares with the actual electricity sales in 2015. The average error of prediction is 1.78. The results show that the method proposed in this paper is reliable, effective and accurate.
【学位授予单位】：华北电力大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM715
[Abstract]:Electricity sales is an important economic assessment index for power grid enterprises. Monthly electricity sales forecasting is an important daily work of power grid enterprise marketing department. Accurate monthly electricity sales prediction can provide marketing decision support for power grid enterprises. It is of great significance to make the plan of increasing supply and expanding sales, to carry out electric energy substitution, to carry out orderly power consumption scheme, and to improve the quality of customer service. At present, the monthly electricity sales forecast of power grid enterprises mostly adopts the methods of comparative analysis, structure analysis, regression analysis, neural network and so on. These methods can be used to predict the electricity sales to a certain extent, but the accuracy of the overall electricity sales prediction of the State Grid Company is not very good. The main reason is that the different characteristics of the electricity sales curves in the various regions of the State Grid Company are not considered. Only one prediction algorithm is used to predict the electricity sales in many areas, which will inevitably lead to the low accuracy of the prediction. In order to solve the above problems, two methods are proposed in this paper. One is the forecasting method of electricity sales based on historical curve. According to the characteristics of the electricity sales curve of the State Grid Company and 27 provincial (municipal) companies in the time domain and the frequency domain, 27 provincial (municipal) companies were clustered. According to the characteristics of the sales curve and the adaptability of the prediction algorithm (SVM regression, BP neural network, ARIMA etc.), the corresponding forecasting methods are selected for different kinds of provincial (municipal) companies. The same prediction algorithm is used for provincial (municipal) companies in the same category. On the basis of forecasting electricity sales based on historical curve, this paper takes weather, economy, holidays and social events into account, and establishes a revised model of electricity sales forecasting based on SVM regression. The forecast accuracy is further improved according to the monthly electricity sales forecast correction model based on the influencing factors. Another method is to take into account the factors of the Spring Festival to adjust electricity sales. Firstly, the method uses the proportion of electricity sales per month in the first quarter of a historical year and the number of days between the first quarter and the Spring Festival in the first quarter to establish a functional relationship. The number of days is input. The ratio of quarter to quarter is output, and the one-variable function is used to forecast the quarterly ratio of electricity sales in March, and then according to the predicted quarterly ratio and the forecast value of electricity sales before adjustment, the forecast value for March is based on the adjustment of Spring Festival factor. Using the forecasting method of this paper, taking the electricity sales data of State Grid Company from 2010 to 2014 as historical data, this paper forecasts the monthly electricity sales of State Grid Company in 2015, and then compares with the actual electricity sales in 2015. The average error of prediction is 1.78. The results show that the method proposed in this paper is reliable, effective and accurate.
【学位授予单位】：华北电力大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM715
【参考文献】
相关期刊论文 前10条
1 颜伟;程超;薛斌;李丹;陈飞;王顺昌;;结合X12乘法模型和ARIMA模型的月售电量预测方法[J];电力系统及其自动化学报;2016年05期
2 李英惠;胥超;;基于时间序列模型的售电量预测方法[J];山东电力技术;2014年06期
3 王伟;杨辉华;刘振丙;李灵巧;;基于极限学习机的短期电力负荷预测[J];计算机仿真;2014年04期
4 王奕萱;;基于统计特征量和支持向量机的短期售电量预测研究[J];华北电力技术;2013年12期
5 潘小辉;刘丽萍;李扬;;提高月度售电量预测精度的一种新方法[J];电力需求侧管理;2013年03期
6 段树乔;段方婕;;基于多项式回归函数的电网公司售电量预测方法[J];数学的实践与认识;2012年18期
7 杨云莹;任玉珑;段锴;;基于虚拟变量与时间序列法的电量需求预测[J];电力需求侧管理;2011年05期
8 曹葵康;沈海斌;;基于多样本的在线支持向量回归算法[J];浙江大学学报(理学版);2011年04期
9 张士强;王雯;王健;;ARIMA模型在城市年用电量预测中的应用[J];电力需求侧管理;2010年06期
10 汪滔;张伟;吕萌明;汪l，
本文编号：2391306
本文链接：https://www.wllwen.com/kejilunwen/dianlidianqilunwen/2391306.html
