基于MTL-SA-LSTM的多元负荷与光伏发电功率短期预测  

Short-term Forecasting of Multivariate Loads and Photovoltaic Power Based on MTL-SA-LSTM

作　　者：张春霞[1] ZHANG Chunxia(Jinan Engineering Polytechnic,Jinan 250200,China)

机构地区：[1]济南工程职业技术学院,山东济南250200

出　　处：《山东电力高等专科学校学报》2024年第6期48-54,共7页Journal of Shandong Electric Power College

摘　　要：在包含光伏发电的综合能源系统中,准确预测多元负荷与光伏发电功率对负荷需求响应计划制定、能源设备调度以及可再生能源消纳至关重要。为此提出一种新型短期预测模型,用于同时预测电负荷、冷负荷、热负荷以及光伏发电功率。该模型采用基于硬参数共享的多任务学习和长短时记忆网络架构,并加入自注意力机制以防止性能下降,模型预测结果与其他模型相比准确性明显提高。It is crucial to accurately predict the multivariate loads and photovoltaic power for the development of load demand re-sponse plan,energy equipment scheduling,and renewable energy consumption in an integrated energy system that includes photo-voltaic power generation.To this end,a novel short-term forecasting model is proposed for the simultaneous forecasting of elec-tric,cooling and thermal multivariate loads as well as photovoltaic power.The model adopts the multi-task learning and long short-term memory network architecture based on hard parameter sharing,with the self-attention mechanism incorporated to pre-vent performance degradation,thus giving significantly higher accuracy in prediction results compared with other models.

关 键 词：自注意力机制 多任务学习 多元负荷 光伏发电功率 短期预测 

分 类 号：TK01[动力工程及工程热物理]