基于MRMR和双重注意力机制的城市能源多元负荷短期预测

Short-term Multivariate Load Forecasting for Urban Energy Based on Minimum Redundancy Maximum Relevance and Dual-Attention Mechanism

为支撑城市能源系统的经济调度和优化运行,将最小冗余最大相关性(MRMR)分析方法与基于双重注意力机制和序列到序列(Seq2Seq)的神经网络相结合,提出一种新型城市能源系统多元负荷短期预测方法。首先,确定目标预测负荷,以MRMR为标准筛选特征序列集,既保持了低冗余度,又保证了输入序列信息的完整性;然后,在Seq2Seq模型基础上,将双重注意力机制融入长短期记忆网络,增强了算法对特征序列时空特征的学习能力;最后,以美国亚利桑那州立大学城市能源系统的实测负荷数据为例进行分析。实验结果表明,所提方法相比现有预测方法具有更高的预测精度和充足的鲁棒性,在4个季节和不同气象误差下都具有良好的表现,可以为城市能源系统的调度运行提供有力的决策依据。

In order to support the economic scheduling and optimal operation of the urban energy system,a new short-term multivariate load forecasting method for the urban energy system is proposed by combining the minimum redundancy maximum correlation(MRMR)analysis method with the neural network based on the dual-attention mechanism and sequence-to-sequence(Seq2Seq).Firstly,the target forecasting load is determined,and MRMR is used as the criterion to screen the characteristic sequence set,which not only maintains low redundancy but also ensures the integrity of input sequence information.Then,based on the Seq2Seq model,the dual-attention mechanism is integrated into the long short-term memory network to enhance the learning ability of the algorithm for the temporal-spatial characteristics of the characteristic sequence.Finally,the measured load data of the urban energy system in Arizona State University,USA is taken as an example for analysis.The experimental results show that compared with the existing forecasting methods,the proposed method has higher forecasting accuracy and sufficient robustness,and performs well in four seasons and with different meteorological errors,which can provide a powerful decision basis for the scheduling operation of the urban energy system.