首页|基于LSTM神经网络的办公建筑逐日能耗预测研究

基于LSTM神经网络的办公建筑逐日能耗预测研究

Daily Energy Consumption Prediction of Office Building Based on LSTM Neural Networks

扫码查看
基于LSTM神经网络建立了办公建筑能耗预测模型,通过引入Adam梯度优化算法自适应地调整学习率,使用2020 年-2022 年上海某办公建筑运营参数和气象参数和能耗数据进行办公建筑逐日能耗预测模型训练,模型的相对误差平均值仅为 4.73%,与其他常用于建筑能耗预测的神经网络算法相比具有较好的准确度.为进一步验证模型的可靠性,避免预测结果断崖式下跌,采用了一段非训练集数据对模型的预测误差进行校核,通过输入 2023 年 1 月至7 月的建筑运营参数和气象参数,对建筑长期能耗预测.实验结果表明,该模型在长期能耗预测中的误差仅为4.26%,可以较为准确地学习序列中的模式和趋势,为办公建筑能耗预测和运营管理提供了实际应用价值.
An office building energy prediction model was established,leveraging the LSTM neural network.The Adam gradient optimization algorithm was introduced to adaptively adjust the learning rate.The model was trained using operational parameters,meteorological parameters,and energy consumption data from 2020 to 2022 for an office building in Shanghai.The average relative error of the model was merely 4.73%.Compared with other commonly used neural network algorithms for building energy consumption prediction,it demonstrates superior accuracy.To ensure the reliability of the model and avoid abrupt declines in prediction results,a set of non-training data was utilized to verify the prediction error.By inputting building operation parameters and meteorological parameters from January to July 2023,the long-term energy consumption of the buildings was predicted.The experimental results indicate that the error of this model in long-term energy consumption prediction is only 4.26%.Consequently,it accurately assimilates patterns and trends within the sequence,thereby providing practical application value for predicting office building energy consumption and managing operations.

LSTMneural networkbuilding energy consumptionautocorrelationenergy consumption prediction

陈家乐、张芸芸、崔红伟

展开 >

上海东方延华节能技术服务股份有限公司,上海 200333

LSTM 神经网络 建筑能耗 自相关性 能耗预测

住房和城乡建设部科技示范项目

S20200064

2024

建筑节能(中英文)
中国建筑东北设计研究院有限公司

建筑节能(中英文)

CSTPCD
影响因子:0.695
ISSN:2096-9422
年,卷(期):2024.52(9)
  • 15