Short-Term Power Load Forecasting Technology Based on Bi-directional LSTM Quantile Regression

Abstract

For the short-term power load probability density forecasting problem of smart grid, a Bi-directional LSTM quantile regression model is proposed. First, a Bi-directional LSTM cell structure is proposed to better extract the high-order features of short-term power load forecasting. Meanwhile, an attention mechanism is introduced into the Bi-directional LSTM structure to capture the importance of each part of the long sequence problem, and then the forecasting accuracy can be improved. Using the Bi-LSTM model, the power load forecast at different quantiles are obtained, and the Gaussian smoothing non-parametric estimation method can used to obtain the power load probability density prediction. Through simulation on actual power load data, the proposed method can provide more accurate short-term power load forecasting performance than previous load probability density prediction methods.