当前位置: X-MOL 学术Energy Build. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Short-term building electricity load forecasting with a hybrid deep learning method
Energy and Buildings ( IF 6.6 ) Pub Date : 2025-01-20 , DOI: 10.1016/j.enbuild.2025.115342
Wenhao Chen, Fei Rong, Chuan Lin

With the increase in energy consumption of the construction industry, building electrical load forecasting (BELF) is of crucial importance. However, there are many research gaps in existing methods: Firstly, existing BELF methods focus on time-domain decomposition, ignoring frequency-domain features and lacking prediction models suitable for different modal components; Secondly, existing models use a single-size convolution kernel, which restricts the extraction of long-term and short-term load characteristics; Thirdly, exogenous variables such as temperature, humidity, and date are not fully considered, resulting in information redundancy. To address these issues, this paper designs a hybrid deep learning model. Firstly, the model classifies the load sequence by adopting the modal component grouping techniques of CEEMDAN decomposition, FFT frequency-domain analysis, and K-means clustering. Subsequently, an ensemble model based on dropout-connected MKDCN-LSTM is designed to learn the nonlinear features of different frequency components. Then, a weight integration strategy based on non-negative constraint theory and the gray wolf optimizer is applied to determine the optimal coefficients. The experimental results indicate that, validated by the historical load data of public buildings, our method can reduce the mean absolute percentage error (MAPE) by up to 1.3% compared with existing ones. It demonstrates the effectiveness of this model in improving the accuracy of BELF.

中文翻译:


使用混合深度学习方法进行短期建筑电力负荷预测



随着建筑业能源消耗的增加,建筑电力负荷预测 (BELF) 变得至关重要。然而,现有方法存在许多研究空白:首先,现有的 BELF 方法侧重于时域分解,忽略了频域特征,缺乏适用于不同模态分量的预测模型;其次,现有模型使用单尺寸卷积核,限制了长期和短期载荷特性的提取;第三,温度、湿度和日期等外生变量没有得到充分考虑,导致信息冗余。为了解决这些问题,本文设计了一个混合深度学习模型。首先,该模型采用 CEEMDAN 分解、FFT 频域分析和 K-means 聚类的模态分量分组技术对载荷序列进行分类;随后,设计了一个基于 dropout-connected MKDCN-LSTM 的集成模型来学习不同频率分量的非线性特征。然后,应用基于非负约束理论和灰狼优化器的权重积分策略来确定最优系数。实验结果表明,通过公共建筑的历史荷载数据验证,与现有方法相比,该方法的平均绝对百分比误差(MAPE)最多可降低1.3%。它证明了该模型在提高 BELF 准确性方面的有效性。
更新日期:2025-01-20
down
wechat
bug