Two-stage household electricity load forecasting based on LSTNet-LSTM

Abstract

To reduce the influence of high volatility and randomness of household load data on short-term load forecasting ,and then fully mine the time-series information of household load data, a two-stage hybrid forecasting algorithm combining long-and short-term time-series network (LSTNet) and long short-term memory network (LSTM) is proposed. In phase I, only the historical load data are used as the input features, LSTNet model is used for deeply mining short-term and long-term time-series information of the load data; In phase II, the prediction errors and multiple influencing factors of LSTNet model are used as the inputs of the LSTM model, the hidden information and periodic attributes of the errors are mined. Experimental validation is carried out on the open-source dataset UCI household electricity dataset, the results show that proposed algorithm has higher forecast accuracy in household load forecasting.