A novel bidirectional recurrent neural network for power load forecasting

In order to effectively improve the accuracy of power load forecasting, a multi-frequency bidirectional short-term power load prediction model combining long short-term memory network, gated cyclic unit and particle swarm algorithm is proposed in view of the characteristics of nonlinearity, non-stationarity and timing of power load. The model first uses K-means clustering analysis to obtain data series with high similarity, and uses variational mode decomposition VMD to decompose and reconstruct the data series to obtain two frequencies. For high-frequency components, the LSTM-GRU bidirectional model is used for prediction; The low-frequency part uses a recurrent neural network that introduces particle swarm arithmetic. Finally, the prediction results obtained by each model are superimposed to obtain the final prediction results. The simulation results show that compared with other network models, the model has high prediction accuracy and fitting ability, and is an effective short-term power load prediction method.