基于AE-TCN的火力发电预测算法研究  

Research on Thermal Power Generation Prediction Algorithm Based on AE-TCN

作　　者：张向阳 郑晓亮[1] ZHANG Xiangyang;ZHENG Xiaoliang(School of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan,Anhui 232001,China)

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232001

出　　处：《自动化应用》2025年第4期15-18,共4页Automation Application

基　　金：国网安徽省电力有限公司科技项目(52120023001Q)。

摘　　要：针对目前火力发电难以预测的问题,提出了一种基于自编码器(AE)和时序卷积网络(TCN)的火力发电预测方法,并与传统的时间序列预测模型长短期记忆(LSTM)网络进行比较。以某火电厂近1年的发电量为研究对象,对日发电量进行预测。通过TCN、LSTM、AE-TCN这3种模型预测发电量数据,通过均方误差、均方根误差和平均绝对误差的比较,证明了基于自编码器和时序卷积网络的时间序列预测模型具有良好的预测精度和预测数据拐点的能力,在火力发电的预测中具有实用性。A thermal power generation prediction method based on Auto Encoder(AE)and Temporal Convolutional Network(TCN)is proposed to address the difficulty of predicting thermal power generation,and compared with traditional time series prediction models such as Long Short-Term Memory(LSTM)network.Using the power generation of a certain thermal power plant in the past year as the research object,predict the daily power generation.By using three models,TCN,LSTM,and AE-TCN,to predict power generation data,and comparing the mean square error,root mean square error,and mean absolute error,it is proven that the time series prediction model based on autoencoder and temporal convolutional network has good prediction accuracy and the ability to predict data inflection points,and is practical in predicting thermal power generation.

关 键 词：火电厂 发电量预测 时序卷积网络 自编码器 

分 类 号：TM611[电气工程—电力系统及自动化] TP39[自动化与计算机技术—计算机应用技术]