基于EMD-MLP组合模型的用电负荷日前预测

Day-Ahead Forecast of Electrical Load Based on EMD-MLP Combination Model

[目的]用电负荷的精准预测是电力系统运行优化的基础,是电力系统能量管理中不可或缺的组成部分。针对传统数据分解技术与机器学习模型结合预测存在的精准度低、计算量大等问题,提出一种将经验模态分解与多层感知机结合(EMD-MLP)的新方法对用电负荷进行日前预测。[方法]首先基于EMD将原始负荷时间序列信号分解为多个本征模函数(Intrinsic Mode Function,IMF)分量,然后采用极值点划分法将多IMF分量进行重构形成高频和低频两个成分以精简预测对象,最后对重构的新分量分别建模预测,并将它们的预测结果叠加作为用电负荷预测值。[结果]采用澳大利亚电力市场2018年、2019年的实测用电负荷数据进行试验。[结论]将建立的EMD-MLP组合模型与持续性模型、单一MLP模型以及传统EMD组合模型进行外推预测效果的对比,验证了所建模型在提高预测精度上的有效性。此外,所提出的EMD-MLP组合新方法在保证精度的同时简化了模型复杂度,提高了预测效率,可以方便地应用于实际中的用电负荷日前与实时预测。

[Introduction]Accurate load forecasting underpins the operation optimization of the electricity systems and is an indispensable aspect of energy management within such systems.Given the low accuracy and high computational complexity inherent in traditional methodologies that combine data decomposition and machine learning models,this study proposes a novel Empirical Mode Decomposition and Multi-Layer Perceptron(EMD-MLP)model for predicting day-ahead electrical load.[Method]Initially,the EMD method decomposed the original load time series into multiple Intrinsic Mode Function(IMF).These IMFs were then reconstructed into high-frequency and low-frequency components using extreme point partitioning,simplifying the prediction target.Subsequently,each reconstructed components was modeled separately for prediction,and the results were cumulatively used to provide the forecasted electrical load value.[Result]The proposed model is tested using real-world electrical load data of 2018 and 2019 from the Australian electricity market.[Conclusion]Comparing the extrapolative capabilities of our EMD-MLP model with persistence model,standalone MLP model and traditional EMD ensemble model confirms the effectiveness of our model in enhancing prediction accuracy.Moreover,while ensuring accuracy,the proposed EMD-MLP model simplifies the complexity and improves the efficiency of the forecasting process,thereby providing a practical solution for both day-ahead and real-time electrical load forecasting.