基于特征选择实现多因素电力负荷预测

Feature-Based Selection for Multi-Factor Power Load Forecasting

负荷预测在保障电网安全运行和提高经济效益方面均占有举足轻重的地位。高精度的负荷预测不仅需要依靠历史负荷数据,并且气象、电价等诸多因素也会对其产生不同程度的影响。为了综合考虑诸多因素造成的影响。首先,以遴选关键因素作为切入点,利用新英格兰地区实测数据集,使用递归特征消除(RFE)有效去除冗余的影响因素,得到与真实负荷相关性高的影响因素,从而构造新的输入负荷数据。然后,利用注意力机制(Attention)动态调整各因素所占权重的特点,提出引入Attention的长短期记忆(LSTM)网络预测模型实现电力负荷预测。试验结果显示,与经典的K近邻方法(KNN)、LSTM、支持向量回归(SVR)等单一算法以及RFE-LSTM等组合算法相比,所提出的RFE-LSTM-Attention方法可以有效地提取关键负荷因素并获得良好的预测性能。

Load forecasting plays a pivotal role in ensuring the safe operation of power grids and improving economic efficiency.High accuracy load forecasting not only relies on historical load data,but also depends on many factors such as weather and electricity prices.To comprehensively consider the multi-factor influence,the key factors are selected as the starting point,and the New England measured data set is used to effectively remove the redundant influencing factors using recursive feature elimination(RFE)to obtain the influencing factors with high correlation with the real load,to construct new input load data.Then,the attention mechanism(Attention)is used to dynamically adjust the weights of each factor.The long short-term memory network(LSTM)prediction model with Attention is proposed to realize the electric load prediction.The experimental results show that the proposed RFE-LSTM-Attention method can effectively extract key load factors and obtain good prediction performance compared with the classical K-nearest neighbor(KNN)method,LSTM,support vector regression(SVR)and other single algorithms as well as RFE-LSTM and other combined algorithms.