IVY Optimized BITCN-BIGRU Model for Power Load Forecasting

With the increasing diversification of the electricity market, power load has exhibited more complex and variable new characteristics. To fully exploit the temporal features in power load data and improve load forecasting accuracy, this paper proposes a power load forecasting model based on IVY-optimized BITCN-BIGRU. First, a bidirectional temporal convolutional network (BITCN) is employed to extract features from the load data, and the extracted temporal features are then input into a bidirectional gated recurrent unit (BIGRU) neural network for temporal prediction. Subsequently, the predicted values are integrated and output through a fully connected layer. Finally, to address the challenge of parameter selection in temporal models, the IVY algorithm is adopted to optimize the model. Case studies are conducted using the 2014 Global Energy Forecasting Competition (GEFCom2014) power load dataset. Simulation results demonstrate that the IVY algorithm achieves rapid convergence in the early stages of iteration, and the constructed IVY-BITCN-BIGRU hybrid forecasting model yields MAE, RMSE, and MAPE values of only 2.7652 MW, 3.8409 MW, and 1.57%, respectively. Compared to the traditional hybrid forecasting model TCN-GRU, these metrics are reduced by 49.99%, 43.64%, and 51.09%, indicating superior forecasting accuracy.