A novel ensemble method for hourly residential electricity consumption forecasting by imaging time series. (15th July 2020)

• Record Type: Journal Article
• Authors: Zhang, Guoqiang; Guo, Jifeng
• Is Part Of: Energy. Volume 203(2020)
• Publication Date: 2020-07-15
• Subjects: Electricity consumption forecasting -- Conditional generative adversarial networks (CGANs) -- Feature transform -- Gramian angular fields (GAFs) -- Improved dragonfly algorithm (IDA); Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.117858
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: In this paper, a novel ensemble method is proposed to forecast the hourly consumption of residential electricity. Firstly, variational mode decomposition (VMD) is applied to decompose weather conditions (relative humidity and temperature, etc.), residential building data (manually operated appliances relevant to residents' lifestyle, dishwasher, heating heat-pump, and television, etc.), and electricity price into several band-limited intrinsic mode functions (BLIMFs). Then the incremental kernel principal component analysis (IKPCA) is applied to extract the incremental kernel principal components (IKPCs) from the BLIMFs. Next, IKPCs are encoded as images by the Gramian Angular Fields (GAFs). Secondly, a novel ensemble method based on conditional generative adversarial networks (CGANs), is applied to simulate the variability in people's electrical behavior and weather forecast errors. Moreover, the elitist search strategy of the multi-population genetic algorithm (MPGA) is introduced to realize the communication among each sub-CGAN. And then all sub-CGANs are integrated by the Huffman coding (HC). Thirdly, an improved dragonfly algorithm (IDA) is developed to optimize the weights of HC. The experimental results show that the forecasting results of the proposed ensemble method are obviously better than those of other standard and state-of-the-art methods tested in this paper. Highlights: Encoding original time series into images. Removing redundant and excessive information. Sample simulation according to input data. Integrated strategy by Huffman coding. Elitist search strategy of the multi-population genetic algorithm.