Attention-based Bi-directional Long and Short Term Memory Networks for Electricity Load Forecasting

With the continuous growth of power load, predicting the trend of power load changes and its efficient management have become increasingly important in recent years. Based on this research, this paper introduces a power load forecasting method based on the Bi LSTM Attention neural network. Firstly, the model only uses univariate data processing to capture short-term changes and long-term trends in the power load time series; Secondly, a deep learning model based on Bi LSTM is constructed to capture hidden features of load time series and introduce attention mechanisms to provide more accurate feature weights; Finally, an improved modified gradient descent algorithm is used for model training and parameter optimization to maximize accuracy. The experimental results show that it’s RMSE and R-squared values reach 8325.4015 and 0.9211, respectively. Compared with BP neural network and a single LSTM network, the model can achieve more accurate prediction results, thus achieving efficient power load management.