Short-term Load Forecasting for Distribution Network Using Decomposition with Ensemble prediction

摘要

short-term load forecasting (STLF) is helpful to grasp the state of the distribution network system. In order to improve the prediction accuracy and robustness, the paper proposes a hybrid model, including Seasonal-Trend decomposition procedure based on Loess (STL), Long Short-Term Memory neural networks (LSTMs) and XGBoost. First, we use the STL method to decompose the load series to mitigate the impact of data distribution on the prediction model. Then, we choose the LSTM neural network as the learner and design an ensemble prediction model to improve the prediction accuracy and robustness. And XGBoost as the last step of ensemble prediction is adopted to combine with all learners to obtain the ensemble prediction results. The proposed STL-LSTMs-XGBoost is applied on the load data collected from 10kV line in Chongqing distribution network. The experimental results can illustrate the feasibility of the model.