Hybrid Multitask Multi-Information Fusion Deep Learning for Household Short-Term Load Forecasting
Lianjie Jiang, Xinli Wang, Wei Li, Lei Wang, Xiaohong Yin, Lei Jia
Abstract
With the detailed load data provided by smart meter, the learning of electricity usage behavior for individual household short-term load forecasting has become a hot research topic. The existing methods generally ignore the relation among multiple households and fail to obtain enough information from the limited household load data, resulting in accuracy reduction. To solve this problem, a novel hybrid <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Multitask Multi-information Fusion Deep Learning Framework</i> (MFDL) considering both recent and long-term regular behaviors is proposed by integrating long short-term memory (LSTM), and convolutional neural network (CNN) to predict the multiple individual household short-term electricity consumptions. A multitask framework with two levels of information extraction is developed considering multiple household short-term load forecasting. In the low-level information extraction stage, the data structure including long and short datasets with different time scales is presented to learn the recent electricity usage behavior by LSTM and the long-term regular electricity behavior features by CNN. Then the multiple time-scale information is fused and further extracted in the high-level stage to complete the multiple individual short-term household load forecasting. The developed framework is implemented on one real-world load datasets and the results show the proposed method outperforms the state-of-the-art methods in common metrics.