Short-term power load forecasting based on combination of residual network and Bi-LSTM

Abstract:

To fully develop the potential characteristics of electric load historical data and improve the prediction accuracy of the short-term load forecasting model, a residual AM-Bi-LSTM prediction model combining improved residual network (ResNetPlus), attention mechanism (AM) and bi-directional long short-term memory network(Bi-LSTM) is proposed in this paper. This model takes historical load, temperature and the features of the predicted date as input. First, based on the Bi-LSTM prediction model, multi-layer improved residual networks are introduced to extract the hidden features of input data, which solves the problem of network degradation caused by the deepening of hidden layers of neural networks, and greatly improves the back propagation ability of the model. Second, the attention mechanism is used to analyze the correlation between input information and current load in the network and highlight the impact of important information, thus improving the speed and accuracy of the model. Third, the Snapshot strategy is used to integrate multiple models that converge on different local minima, in order to improve the accuracy and robustness of the model. Finally, the US ISO-NE Dataset is used to verify the performance of the model. The experimental results show that the proposed model has achieved an average prediction accuracy of 98.27%. The average prediction accuracy over 12 consecutive months with the proposed model has improved by 2.87% compared to the traditional LSTM model. In addition, the average prediction accuracy under different seasons based on the proposed model has improved by 1.03% and 1.16% compared to the AM-Bi-LSTM and ResNetPlus models, respectively. This indicates that compared to the contrast model, the residual AM-Bi-LSTM model has higher accuracy, robustness and generalization ability.