基于特征降维和组合模型的短期电力负荷预测  

Short-Term Power Load Forecasting Based on Dimensionality Reduction and Combined Model

作　　者：徐先峰[1] 赵依 龚美 陈雨露 XU Xian-feng;ZHAO Yi;GONG Mei;CHEN Yu-lu(College of Electronics and Control Engineering,Chang'an University,Xian Shanxi 710064,China)

机构地区：[1]长安大学电子与控制工程学院,陕西西安710064

出　　处：《计算机仿真》2022年第4期66-70,230,共6页Computer Simulation

基　　金：国家自然科学基金(61201407,71971029);陕西省自然科学基础研究计划(2016JQ5103,2019GY-002);长安大学中央高校基本科研业务费(300102328202);西安市智慧高速公路信息融合与控制重点实验室(ZD13CG46)。

摘　　要：准确的负荷预测是电力系统安全稳定运行的重要保障,为了进一步提高电力负荷的短期预测精度,依托信号处理和深度学习技术,针对电力负荷数据的特征降维方法以及组合模型的构建进行深入研究。首先利用随机森林的平均不纯度减少法(MDI)实现多维负荷数据的特征降维,实验结果证明上述方法能有效筛选出影响负荷的主要因素,提高模型学习效率。在此基础上,提出融合多种算法优点的CEEMDAN-ARIMA-LSTM组合模型,通过设置上述组合模型与单一LSTM模型以及AutoEncode-VMD-BP的对比实验,有力论证了所提模型在负荷预测方面具有更高的精确度和适用性。Accurate load forecasting is an important guarantee for the safe and stable operation of power systems.In order to further improve the short-term forecasting accuracy of power load, relying on signal processing and deep learning technology, the feature dimensionality reduction method of power load data and the construction of combined model were deeply studied. Firstly, Mean Decrease Impurity(MDI) of Random Forest was used to achieve the feature dimension reduction of multi-dimensional load data. The experimental result shows that the method can effectively screen out the main factors affecting the load and improve the model learning efficiency. On this basis, CEEMDANARIMA-LSTM model combined with the advantages of multiple algorithms was proposed. By setting the comparison model of the combination model with LSTM network and Auto Encode-VMD-BP model, it strongly demonstrates that the proposed model in load forecasting has higher accuracy and applicability.

关 键 词：信号处理 深度学习 长短时记忆网络 特征降维 组合模型 

分 类 号：TM743[电气工程—电力系统及自动化]