基于多视角时间特征的短期电力负荷预测
Short-term Power Load Forecasting Based on Multi-view Temporal Features
李波 1罗清 2高翔 2陈祖秀 2胡屹立2
作者信息
- 1. 云南电网有限责任公司电力科学研究院,云南 昆明 650100
- 2. 云南电网有限责任公司玉溪供电局,云南 玉溪 653100
- 折叠
摘要
外部因素对电力负荷具有周期性影响,且这些影响直接反映在电力负荷值上.基于多视角表示学习思想,使用历史电力负荷预测值的不同视角作为外部因素的隐藏表示.通过对历史电力负荷数据进行特征提取,并将电力负荷分为分钟、小时和天三个时间视角,分别采用了适应性的神经网络模型进行特征提取,并引入了一个多视角特征合并模块,融合不同时间尺度上的信息来提高负荷预测准确性.实验证明,所提出的方法在西南某地区的电力负荷数据集上表现出较好的预测性能,与单一时间视角的模型相比,平均绝对误差和均方误差分别降低了12.21%和11.12%.
Abstract
External factors exhibit periodic influences on power load,and these impacts are directly reflected in load val-ues.Drawing inspiration from the concept of multi-perspective representation learning,this study leverages various per-spectives of historical load forecasting values as hidden representations of external factors.By extracting features from his-torical electricity load data and dividing the load into three time perspectives-minutes,hours,and days-tailored neural net-work models are employed for feature extraction based on these different time perspectives.Furthermore,a multi-per-spective feature fusion module is introduced,amalgamating information across different time scales to enhance load fore-casting accuracy.Empirical results demonstrate that the proposed method exhibits superior predictive performance on a power load dataset from a specific region in Southwest China.In comparison to models solely considering a single time per-spective,the proposed approach achieved reductions in MAE and MSE of 12.21%and 11.12%,respectively.
关键词
短期电力负荷预测/多视角时间特征/多特征融合Key words
short-term power load forecasting/multi-view temporal feature/multi-feature fusion引用本文复制引用
出版年
2024
NETL
NSTL
万方数据