以直供用电量预测经营区域及全国用电量的方法研究  

Regional and National Electricity Consumption Forecast Based on State Grid Direct Power Supply

作　　者：胡航海[1] 刘道新[1] 张健[1] 李鸿琪 

机构地区：[1]国家电网公司,北京100031 [2]北京中电普华信息技术有限公司,北京100192

出　　处：《华东电力》2014年第5期1023-1026,共4页East China Electric Power

摘　　要：以国家电网直供用电量精准地(相对误差在0.5%以内)预测国家电网经营区域及全国用电量,可以为各级领导宏观经济决策做好辅助工作。大量数据的测算证明,没有一种方法可以保证一年12月中每月都达到精准预测,故使用了多种预测方法,其中以RBF神经网络和带虚拟变量回归方法为主要预测方法,X-12一Arima、指数平滑和季节解构方法作为辅助方法,通过不同方法的互为补充,找到每个月度预测的最好方法,并在此基础上建立了预测模板。利用模板信息,根据2013年11月经营区域直供用电量对经营区域及全国用电量进行了预测,预测结果与实际用电量对照,经营区域用电量相对误差为0.16%,全国用电量相对误差为0.48%。It can provide assistance to leaders at all levels in macroeconomic decision-making to forecast regional and national electricity consumption based on state grid direct power supply accurately(the relative error is within 0.5%).A large amount of data calculation has verified that it is impossible to find one method to guarantee accurate forecasts every month in a year,so this paper employs a variety of prediction methods,particularly RBF neural network and virtual variable regression method as the primary methods,and X-12-Arima,exponential smoothing and season deconstruction method as auxiliary methods.Through using the various and complementary methods,the best approach to find monthly forecast is defined.Thereupon the prediction model is established.The model information is adopted for the regional and national power consumption prediction based on the direct power supply in November2013.The comparison between the actual power consumption and the prediction results indicates that the relative error of regional and national power consumption is 0.16%and 0.48%,respectively.

关 键 词：RBF神经网络 回归 X-12-ARIMA 指数平滑 季节解构 

分 类 号：F426.61[经济管理—产业经济] F224