大数据技术在电网数据上的应用  

Application of big data technology in state grid data

作　　者：涂炼 张水平 左剑 李顺 方海泉 鲍威[2] TU Lian;ZHANG Shuiping;ZUO Jian;LI Shun;FANG Haiquan;BAO Wei(Electric Power Dispatching and Control Center of Guangdong Power Grid Company Limited,Guangzhou 510030,China;College of Electrical Engineering,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]广东电网有限责任公司电力调度控制中心,广东广州510030 [2]浙江大学电气工程学院,浙江杭州310058

出　　处：《电气应用》2021年第10期52-57,共6页Electrotechnical Application

基　　金：南方电网公司科技项目资助(GDKJXM20190037)。

摘　　要：为了有效预测电厂机组出力情况,不仅要采用先进的预测方法,还要对机组出力数据进行合理的预处理。提出了电厂机组出力数据分析的一整套流程,包括数据集成、异常值预处理、数据可视化和电厂机组出力预测。以某省2018年机组出力数据为研究对象,选取一个燃煤电厂为例。经过数据集成、数据预处理和可视化展示,并用长短期记忆(LSTM)神经网络对电厂出力进行预测,LSTM预测得到的平均绝对百分比误差(MAPE)为10.90%,预测结果优于误差反向传播(BP)神经网络,BP神经网络预测得到的MAPE为11.61%。说明经过预处理的机组出力数据再用LSTM模型预测能达到良好的预测准确度。In order to effectively predict the output of power plant units, it is necessary not only to adopt advanced prediction methods, but also to reasonably preprocess the unit output data. A complete set of process of power plant unit output data analysis is proposed, including data integration, abnormal value preprocessing, data visualization and power plant output prediction, which can provide reference for power decision-making. We take the unit output data of Certain Province in 2018 as the research object, and take the coal-fired power plant as an example. After data integration, data preprocessing and visualization, the power plant output is predicted by using long short term memory(LSTM) neural network, and the mean absolute percentage error(MAPE) is 10.90%. The prediction result is better than the error back-propagation(BP) neural network, which MAPE is 11.61%. It is proved that the preprocessing unit output data can achieve good prediction accuracy by using LSTM model.

关 键 词：电厂 大数据 异常值 可视化 长短期记忆神经网络 预测 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论] TM73[自动化与计算机技术—计算机科学与技术]