首页|基于迁移深度强化学习的火电机组实时碳排放预测方法

基于迁移深度强化学习的火电机组实时碳排放预测方法

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
由于火电厂机组碳数据类型多样、样本过多,导致机组碳排放预测精度低,为此提出了基于迁移深度强化学习的火电机组实时碳排放预测方法.首先,设计火电厂碳排放数据自动采集系统进行数据采集;其次,引入神经网络的全自动编码器对样本数据解码重构,结合深度迁移学习的方法提取碳排放数据特征;再次,设计强化学习的支持向量机回归模型,把经过特征提取后的数据导入线性回归函数中,进行非线性回归计算;最后,计算样本集的拟合值和均方误差值,把碳排放影响因素经过迁移深度强化学习归一化处理后带到预测模型进行计算,实现火电机组碳排放实时预测.由实验可知,所提方法对2022年测试机组碳排放量预测的最大误差为0.02 × 104 t,均方误差和平均绝对误差较小,预测精准度高,具有较好的应用效果.
Real Time Carbon Emission Prediction Method for Thermal Power Units Based on Transfer Deep Reinforcement Learning
Due to the diversity of carbon data types and excessive sample size in thermal power plant units,the accuracy of carbon emission prediction is low.Therefore,a real-time carbon emission prediction method for thermal power units based on transfer deep reinforcement learn-ing is proposed.Firstly,design an automatic data collection system for carbon emissions in thermal power plants for data collection;Secondly,a fully automatic encoder based on neural networks is introduced to decode and reconstruct the sample data,and deep transfer learning is used to extract the features of carbon emissions data;Once again,design a support vector machine regression model for reinforcement learning,im-port the feature extracted data into a linear regression function,and perform nonlinear regression calculations;Finally,the fitting values and mean square error values of the sample set are calculated,and the factors affecting carbon emissions are normalized through deep transfer rein-forcement learning and brought into the prediction model for calculation,achieving real-time prediction of carbon emissions from thermal power units.According to the experiment,the maximum error of the proposed method in predicting the carbon emissions of test units in 2022 is 0.02 × 104 tons,with small mean square error and average absolute error,high prediction accuracy,and good application effect.

thermal power plantcarbon emission predictionfeature extractiontransfer deep learningreinforcement learningsupport vec-tor regression machine

袁鹏、谭琛、李锋、张银芽、唐述

展开 >

国家电网有限公司华中分部,湖北 武汉 450001

北京科东电力控制系统有限责任公司,北京 100192

重庆邮电大学计算机科学与技术学院,重庆 400065

火电厂 碳排放预测 特征提取 迁移深度学习 强化学习 支持向量回归机

国家自然科学基金项目重庆市技术创新与应用发展专项面上项目

61601070cstc2020jscxmsxmX0135

2024

10.3969/j.issn.1002-1639.2024.07.013

工业加热
西安电炉研究所有限公司

工业加热

CSTPCD
影响因子：0.257
ISSN：1002-1639
年,卷(期)：2024.53(7)
  • 15