摘要
针对电力企业碳排放核算时间长、连续排放监测系统误差大及传统模型拟合困难等问题,结合电力企业燃料燃烧的特性及现有污染物在线监测结果,成功构建了电力行业碳排放的残差双向长短期记忆效应网络(ResNet-BiLSTM)模型,并以浙江省113家电力企业的数据为样本进行验证。结果表明:与目前主流数据预测算法逻辑回归(Regression)、循环神经网络(RNN)、反向传播神经网络(BPNN)模型相比,ResNet-BiLSTM模型的平均绝对百分比误差分别低5.7、4.1、2.8百分点,对碳排放量的预测更贴近电力企业核算碳排放波动情况,且预测准确率(96%)最高。ResNet-BiLSTM模型的成功应用不仅为电力企业提供了新的碳排放预测途径,同时为提高相关管理部门的碳排放数据监管效率提供了支持。
In response to the problems of long carbon emission accounting time,large errors in continuous emission monitoring systems,and difficulties in fitting traditional models in power companies,the ResNet-BiLSTM model for carbon emissions in the power industry was successfully constructed,which combined with the characteristics of fuel combustion in power companies and the existing online monitoring results of pollutants.The model was validated using data from 113 power companies in Zhejiang Province as samples.The results showed that compared with current mainstream data prediction algorithms such as Regression,RNN and BPNN models,the average absolute percentage error of ResNet-BiLSTM model was 5.7,4.1,and 2.8 percentage points lower,respectively.The prediction of carbon emissions was closer to the calculation of carbon emissions fluctuations by power companies,and the prediction accuracy(96%)was the highest.The successful application of the ResNet-BiLSTM model not only provided a new approach for carbon emission prediction for power companies,but also supported the improvement of carbon emission data supervision efficiency for relevant management departments.
作者
陈齐
许明海
沈赛燕
郭磊
CHEN Qi;XU Minghai;SHEN Saiyan;GUO Lei(Zhejiang Ecological Environment Group Co.,Ltd.,Hangzhou Zhejiang 311100;Zhejiang Tianfeng Environmental Technology Co.,Ltd.,Hangzhou Zhejiang 310012;Zhejiang Jinxin Construction Engineering Co.,Ltd.,Hangzhou Zhejiang 311300)
出处
《环境污染与防治》
CAS
CSCD
北大核心
2024年第5期689-693,720,共6页
Environmental Pollution & Control
作者简介
第一作者:陈齐,男,1977年生,本科,高级工程师,主要从事应对气候变化、环境经济等方面研究。
