基于Dropout-LSTM模型的城市燃气日负荷预测  被引量:4

Daily Load Forecasting of Urban Gas Based on Dropout-LSTM Model

在线阅读下载全文

作  者:于铭多 郝学军[1] YU Mingduo;HAO Xuejun

机构地区:[1]北京建筑大学环境与能源应用工程学院,北京100044

出  处:《煤气与热力》2023年第2期V0010-V0016,共7页Gas & Heat

摘  要:提出采用Dropout技术的长短期记忆神经网络模型(Dropout-LSTM模型),对城市燃气日负荷进行预测。由于不同时期的燃气日负荷具有不同特点,将全年分为供暖期、过渡期及非供暖期,分别对3个时期的日负荷和影响因素进行相关性分析,确定3个模型的输入特征,建立3个时期的日负荷预测Dropout-LSTM模型,采用平均绝对百分比误差对模型预测效果进行评价。Dropout-LSTM模型可以很好地预测城市燃气日负荷,比BP模型、LSTM模型以及SVM模型有更好的预测效果。与基于全年数据的全年预测模型相比,分时期预测模型预测精度更高。供暖期的燃气日负荷规律性强,对供暖期的日负荷预测精度最高,非供暖期次之,由于过渡期日负荷波动大,预测效果是3个时期中最差的。Long Short-Term Memory(LSTM) neural network model using Dropout technology(Dropout-LSTM model) is proposed to predict the daily load of urban gas. Since the daily gas load in different periods has different characteristics, the whole year is divided into three parts: heating period, transition period and non-heating period. The correlation analysis of the daily load and influencing factors of the three periods is carried out to determine the input features of the three models. The Dropout-LSTM models of daily load forecasting for three periods are established, and the average absolute percentage error is used to evaluate the prediction effect of the model. The Dropout-LSTM models can predict the daily load of city gas very well and have better forecasting effect than BP model, LSTM model and SVM model. Compared with the annual forecasting model based on annual data, the forecasting accuracy of the phased forecasting models is higher. The daily gas load in the heating period has strong regularity, and the forecasting accuracy of daily load in the heating period is the highest, followed by non-heating period. Due to the large fluctuation of daily load during the transition period, the forecasting effect is the worst among the three periods.

关 键 词:燃气日负荷 负荷预测 Dropout-LSTM模型 相关性分析 预测精度 

分 类 号:TU996.3[建筑科学—供热、供燃气、通风及空调工程]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象