基于二次分解重构与多任务学习的综合能源系统多元负荷短期预测  

Multi-Energy Load Forecasting of Integrated Energy System based on Secondary Decomposition-Reconstruction and Multi-Task Learning

作　　者：于润泽 窦震海[1] 张志一 胡亚春 陈佳佳[1] 尹文良 YU Runze;DOU Zhenhai;ZHANG Zhiyi;HU Yachun;CHEN Jiajia;YIN Wenliang(School of Electrical and Electronic Engineering,Shandong University of Technology,Zibo 255000,Shandong Province,China)

机构地区：[1]山东理工大学电气与电子工程学院,山东省淄博市255000

出　　处：《电力建设》2024年第12期149-161,共13页Electric Power Construction

基　　金：国家自然科学基金项目(52005306);山东省自然科学基金项目(ZR2020QE220)。

摘　　要：在综合能源系统中,多能源负荷的物理特性差异和耦合特性导致其波动性和随机性显著增强,这使得预测精度难以提高。对此,提出了基于二次分解重构与多任务学习的混合预测模型。针对负荷波动的大量噪声信号,提出二次分解重构策略,通过变分模态分解与改进的自适应噪声完备集合经验模态分解将负荷数据分解为高中低三个频段,再利用排列熵筛选出更能反映负荷变化特征的低频和中频序列。针对多能源负荷的强耦合问题,采用能够捕捉多输出相关信息的多输出最小二乘支持向量回归算法,并结合基于多任务学习的双向长短期记忆网络,分别完成低、中频分量预测。仿真结果显示,二次分解重构数据处理方法大大提升了模型的预测精度,展现了多输出最小二乘支持向量回归在多元负荷预测中的潜在优势。In the field of integrated energy systems,the inherent variability and interconnected nature of various energy loads substantially amplify their unpredictability,posing challenges to enhancing forecast precision.This study introduces a hybrid forecasting model that utilizes quadratic decomposition reconstruction coupled with multitask learning to tackle this.Addressing the prevalent noise in load fluctuations,the model employs a quadratic decomposition strategy.The model leverages variational mode decomposition alongside an improved adaptive noise complete ensemble empirical mode decomposition,segmenting load data into distinct high,medium,and low-frequency bands.Subsequently,permutation entropy is utilized to identify low-and medium-frequency sequences,which more accurately mirror the dynamics of load changes.The model incorporates a multi-output least squares support vector regression algorithm to manage the intricate interdependencies of multiple energy loads.This algorithm excels in assimilating multi-output related data and,in conjunction with a bidirectional long short-term memory network founded on multitask learning,it forecasts the low-and medium-frequency components.Empirical simulations validate that the quadratic decomposition reconstruction approach significantly elevates the predictive accuracy of the model.Additionally,these simulations showcase the prospective benefits of employing multi-output least squares support vector regression for complex multielement load forecasting.

关 键 词：综合能源系统 多元负荷预测 分解重构 多任务学习 多输出最小二乘支持向量回归 

分 类 号：TM73[电气工程—电力系统及自动化]