| 基于气象因子及机器学习回归算法的夏季空调负荷预测 |
| |
| 引用本文： | 田心如,蔡凝昊,张志薇.基于气象因子及机器学习回归算法的夏季空调负荷预测[J].气象科学,2019,39(4):548-555. |
| |
| 作者姓名： | 田心如 蔡凝昊 张志薇 |
| |
| 作者单位： | 江苏省气象台, 南京 210008,江苏省气象台, 南京 210008,中国气象局交通气象重点开放实验室, 南京 210008;江苏省气象科学研究所, 南京 210009 |
| |
| 基金项目： | 国家自然科学基金青年基金资助项目（41805036）；中国气象局2017年决策气象服务专项；江苏省气象局科研面上项目（KM201708） |
| |
| 摘 要： | 气温、气压、相对湿度等气象因子对夏季用电负荷的影响非常显著。为了定量研究气象因子导致用电负荷的变化,本文将夏季用电负荷与当年4月及9月用电平均值之差定义为夏季空调负荷,并利用2014年1月到2016年12月南京市逐时气温、气压、相对湿度、水汽压、降雨量、风速、露点温度等气象资料,以及逐日逐时用电负荷数据资料,采用多元线性、K近邻法,决策树,bagging回归、随机森林等5种机器学习回归算法进行建模,并对其分别进行参数调优工作,进而得到空调负荷预测结果。结果表明:多元线性回归方法是5种回归算法里效果最差的一种,但通过增加特征量的种类和样本数,可以提高预测精度;随机森林回归算法是5种回归算法里效果最好的一种,较多元线性回归算法减小误差达44%,并且较好描述了空调负荷高值区的极端情况并减少了对于训练数据的过拟合现象。
|
| 关 键 词： | 机器学习 回归 随机森林 空调负荷 预测 |
| 收稿时间： | 2018/10/25 0:00:00 |
| 修稿时间： | 2019/1/23 0:00:00 |
Summer air-conditioning load forecasting in Nanjing based on meteorologicalfactors and machine learning regression algorithm |
| |
| TIAN Xinru,CAI Ninghao and ZHANG Zhiwei.Summer air-conditioning load forecasting in Nanjing based on meteorologicalfactors and machine learning regression algorithm[J].Scientia Meteorologica Sinica,2019,39(4):548-555. |
| |
| Authors: | TIAN Xinru CAI Ninghao and ZHANG Zhiwei |
| |
| Institution: | Jiangsu Meteorological Observatory, Nanjing 210008, China,Jiangsu Meteorological Observatory, Nanjing 210008, China and Key Laboratory of Transportation Meteorology, China Meteorological Administration, Nanjing 210008, China;Jiangsu Institute of Meteorological Sciences, Nanjing 210009, China |
| |
| Abstract: | The meteorological factors such as temperature, air pressure and relative humidity have a significant impact on summer electricity load. In order to quantitatively study the change of electricity load caused by meteorological factors, the summer air-conditioning load is defined as the difference between the summer electricity load and the average electricity consumption in April and September of the year. The meteorological data in Nanjing from January 2014 to December 2016, consisting of the hourly temperature, air pressure, relative humidity, water vapor pressure, rainfall, wind speed, dew temperature, and the hourly daily electricity load data, are used to forecast air conditioning load by using five machine regression algorithms including multivariate linear, K-Nearest neighbor, decision tree, bagging regression and random forest for modeling and optimizing the parameters, respectively. The results show that the multiple linear regression method performs worst among the five machine regression algorithms, but increasing the number of parameters'' types and samples can improve the forecast accuracy. Random forest regression algorithm is the most effective method of the five machine regression algorithms, which reduces the error rate by 44% compared with the multiple linear regression, better describes the extreme conditions of high air-conditioning load and reduces the overfitting phenomenon of the training data. |
| |
| Keywords: | machine learning regression random forest air-conditioning load forecast |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《气象科学》浏览原始摘要信息 |
| 点击此处可从《气象科学》下载免费的PDF全文 |
|
