Short-Term Electrical Load Forecasting Based on Whale Optimized Variational Mode Decomposition and DCNN-BiLSTM

Short-term electricity load forecasting has a significant impact on building electricity IoT systems. To deal with the non-linearity and non-smoothness of electrical loads and to tap the influence of multi-dimensional features on electrical loads, this paper proposes a DCNN-BiLSTM network combining Convolutional (CNN), Double Attention (DA) and Bi-directional Long Short Term Memory Network (BiLSTM), which is applied together with Whale Optimisation Variational Modal Decomposition (WOA-VMD) for short-term electrical load forecasting (referred to as WVDCB model). The raw load data is firstly decomposed into less noisy components by the Whale Optimisation VMD algorithm, and secondly the components are predicted by the DCNN-BiLSTM network, which further explores the correlation between the load and multiple features and the key time nodes. Finally, the empirical analysis of the measured data from the UK grid in 2022, comparing the mean absolute error (MAE) equivalent, shows that the WVDCB model has higher accuracy compared to the traditional deep learning model and other hybrid modal decomposition and deep learning models, validating the superiority of the WVDCB forecasting model.